fix(firecracker): honor cached image policy

- config_store.py imported host_db_path from supervise_types (wrong);
  it lives in paths.py, matching all other stores (audit, queue, etc.)
- test_stale_checks: two tests expected check_stale called twice
  (agent + sidecar) — consolidated arch has no sidecar, so once is
  correct; update assertions and remove unused Path import

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitea/workflows/lint.yml

@@ -24,8 +24,19 @@ jobs:
 
       - name: Run pylint
         run: |
-          # Run pylint on all Python files in the repo
-          find . -name '*.py' -not -path './.venv/*' -not -path './.git/*' | xargs pylint --fail-under=8.0
+          # Pylint's normal exit code is nonzero for any emitted finding,
+          # regardless of --fail-under. Preserve the full report but enforce
+          # the aggregate score this workflow promises.
+          set +e
+          find . -name '*.py' -not -path './.venv/*' -not -path './.git/*' \
+            | xargs pylint --fail-under=8.0 \
+            | tee /tmp/pylint-output.txt
+          set -e
+          SCORE=$(sed -n \
+            's/^Your code has been rated at \([-0-9.]*\)\/10.*/\1/p' \
+            /tmp/pylint-output.txt | tail -1)
+          test -n "$SCORE"
+          awk -v score="$SCORE" 'BEGIN { exit !(score >= 8.0) }'
 
       - name: Run pyright
         run: |

.gitea/workflows/tracker-policy-issues.yml

@@ -0,0 +1,17 @@
+name: tracker-policy-issues
+
+on:
+  issues:
+    types: [opened, unlabeled]
+
+jobs:
+  label-issue:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+    steps:
+      - uses: actions/checkout@v4
+      - name: Ensure the issue has a label
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: python3 scripts/tracker_policy.py label-issue

.gitea/workflows/tracker-policy-pr.yml

@@ -0,0 +1,18 @@
+name: tracker-policy-pr
+
+on:
+  pull_request:
+    types: [opened, edited, reopened, synchronized, labeled, unlabeled]
+
+jobs:
+  check-pr:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: read
+      pull-requests: read
+    steps:
+      - uses: actions/checkout@v4
+      - name: Require an unlabeled PR linked to an issue
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: python3 scripts/tracker_policy.py check-pr

README.md

@@ -171,6 +171,15 @@ When an outbound DLP detector matches a token, the route's `dlp.outbound_on_matc
 
 More examples in `examples/`. Full design lives under `docs/prds/`; the trust-boundary rationale is in `docs/prds/0011-per-file-md-manifest.md`.
 
+## Tracker policy
+
+Issues are the canonical work items and own all tracker labels; every issue
+must have at least one. Pull requests stay unlabeled and deliberately reference
+an issue with `Closes #…`, `Part of #…`, or another form defined in
+[`ADR 0005`](docs/decisions/0005-issues-own-tracker-metadata.md). Gitea Actions
+enforces the convention for new work from 2026-07-18 onward. Earlier closed
+PRs are grandfathered rather than given artificial retrospective issues.
+
 ## Trademarks
 
 bot-bottle is an independent project and is not affiliated with, endorsed by, or sponsored by Anthropic, PBC. "Claude" and "Claude Code" are trademarks of Anthropic, PBC; the project name uses "claude" descriptively to indicate that the tool runs Claude Code inside a sandbox.

bot_bottle/backend/__init__.py

@@ -37,10 +37,10 @@ import os
 import shlex
 import sys
 from abc import ABC, abstractmethod
-from contextlib import AbstractContextManager
+from contextlib import AbstractContextManager, contextmanager
 from dataclasses import dataclass
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, Generic, Sequence, TypeVar
+from typing import TYPE_CHECKING, Any, Generator, Generic, Sequence, TypeVar
 
 from ..agent_provider import AgentProvisionPlan, get_provider, build_agent_provision_plan
 from ..egress import EgressPlan
@@ -82,6 +82,9 @@ class BottleSpec:
     # True when launched via --headless (no TTY, no interactive prompts).
     # The git-gate host-key preflight uses this to error rather than prompt.
     headless: bool = False
+    # Image startup policy. "fresh" preserves the normal build path;
+    # "cached" reuses the current local image/artifact without rebuilding.
+    image_policy: str = "fresh"
 
 
 @dataclass(frozen=True)
@@ -277,6 +280,18 @@ PlanT = TypeVar("PlanT", bound=BottlePlan)
 CleanupT = TypeVar("CleanupT", bound=BottleCleanupPlan)
 
 
+@dataclass(frozen=True)
+class BottleImages:
+    """Resolved image references (or artifact paths) for a bottle launch.
+
+    For Docker/macOS-container backends, `agent` and `sidecar` are string
+    image refs. For the smolmachines backend they are Path objects pointing
+    to pre-built `.smolmachine` artifacts."""
+
+    agent: str | Path
+    sidecar: str | Path = ""
+
+
 class BottleBackend(ABC, Generic[PlanT, CleanupT]):
     """Abstract base for selectable bottle backends. Concrete subclasses
     (e.g. DockerBottleBackend) own their own prepare/launch impls.
@@ -436,9 +451,27 @@ class BottleBackend(ABC, Generic[PlanT, CleanupT]):
         prompt file, Dockerfile path, and guest home all live on
         `agent_provision_plan` — the source of truth."""
 
+    def prelaunch_checks(self, plan: PlanT) -> None:
+        """Raise StaleImageError if any cached image used by this plan is stale.
+        No-op default; backends override to call the shared check_stale*
+        helpers on their image/artifact timestamps. Called by the CLI before
+        launch so the operator can be prompted outside the launch context."""
+
+    @contextmanager
+    def launch(self, plan: PlanT) -> Generator[Bottle, None, None]:
+        """Template: build or load images, then delegate to _launch_impl."""
+        images = self._build_or_load_images(plan)
+        with self._launch_impl(plan, images) as bottle:
+            yield bottle
+
     @abstractmethod
-    def launch(self, plan: PlanT) -> AbstractContextManager[Bottle]:
-        """Build/run the bottle and yield a handle; tear down on exit."""
+    def _build_or_load_images(self, plan: PlanT) -> BottleImages:
+        """Return the agent and sidecar image references (or artifact paths)
+        for this plan, building fresh images when the policy requires it."""
+
+    @abstractmethod
+    def _launch_impl(self, plan: PlanT, images: BottleImages) -> AbstractContextManager[Bottle]:
+        """Bring up the bottle using pre-resolved images; yield a handle; tear down on exit."""
 
     def provision(self, plan: PlanT, bottle: "Bottle") -> str | None:
         """Copy host-side files (CA cert, prompt, skills, .git) into

bot_bottle/backend/docker/backend.py

@@ -31,7 +31,7 @@ from ...env import ResolvedEnv
 from ...git_gate import GitGatePlan
 from ...supervise import SupervisePlan
 from ...manifest import Manifest
-from .. import ActiveAgent, BottleBackend, BottleSpec
+from .. import ActiveAgent, BottleBackend, BottleImages, BottleSpec
 from . import cleanup as _cleanup
 from . import enumerate as _enumerate
 from . import launch as _launch
@@ -100,9 +100,15 @@ class DockerBottleBackend(BottleBackend["DockerBottlePlan", "DockerBottleCleanup
             stage_dir=stage_dir,
         )
 
+    def prelaunch_checks(self, plan: DockerBottlePlan) -> None:
+        _launch.stale_checks(plan)
+
+    def _build_or_load_images(self, plan: DockerBottlePlan) -> BottleImages:
+        return _launch.build_or_load_images(plan)
+
     @contextmanager
-    def launch(self, plan: DockerBottlePlan) -> Generator[DockerBottle, None, None]:
-        with _launch.launch(plan, provision=self.provision) as bottle:
+    def _launch_impl(self, plan: DockerBottlePlan, images: BottleImages) -> Generator[DockerBottle, None, None]:
+        with _launch.launch(plan, images, provision=self.provision) as bottle:
             yield bottle
 
     def ensure_orchestrator(self) -> str:

bot_bottle/backend/docker/launch.py

@@ -42,7 +42,9 @@ from ...git_gate import (
     provision_git_gate_dynamic_keys,
     revoke_git_gate_provisioned_keys,
 )
-from ...log import info, warn
+from ...image_cache import check_stale
+from ...log import die, info, warn
+from .. import BottleImages
 from . import util as docker_mod
 from .bottle import DockerBottle
 from .bottle_plan import DockerBottlePlan
@@ -70,16 +72,47 @@ from ...orchestrator.gateway import DockerGateway
 _REPO_DIR = str(Path(__file__).resolve().parent.parent.parent.parent)
 
 
+def build_or_load_images(plan: DockerBottlePlan) -> BottleImages:
+    """Resolve the agent image ref for this plan.
+
+    Returns the committed snapshot if one exists, the cached image when the
+    policy is 'cached', or builds a fresh image and returns that."""
+    committed = read_committed_image(plan.slug)
+    if committed and docker_mod.image_exists(committed):
+        info(f"using committed image {committed!r}")
+        return BottleImages(agent=committed)
+    if plan.spec.image_policy == "cached":
+        if not docker_mod.image_exists(plan.image):
+            die(
+                f"cached agent image {plan.image!r} not found; "
+                "run without --cached-images to build it"
+            )
+        info(f"using cached agent image {plan.image!r}")
+        return BottleImages(agent=plan.image)
+    docker_mod.build_image(plan.image, _REPO_DIR, dockerfile=plan.dockerfile_path)
+    docker_mod.verify_agent_image(
+        plan.image, runtime_for(plan.agent_provider_template).smoke_test,
+    )
+    return BottleImages(agent=plan.image)
+
+
 @contextmanager
 def launch(
     plan: DockerBottlePlan,
+    images: BottleImages,
     *,
     provision: Callable[[DockerBottlePlan, "DockerBottle"], str | None],
 ) -> Generator[DockerBottle, None, None]:
-    """Build, launch, and provision a Docker bottle via compose.
-    Teardown on exit."""
+    """Launch and provision a Docker bottle via compose. Teardown on exit."""
     stack = ExitStack()
 
+    # Stamp the resolved agent image ref into the plan so compose rendering
+    # picks up the right image (may be a committed snapshot or cached ref).
+    plan = dataclasses.replace(
+        plan,
+        agent_provision=dataclasses.replace(plan.agent_provision, image=str(images.agent)),
+    )
+
     _bottle_for_revoke = plan.manifest.bottle
     _git_gate_dir_for_revoke = git_gate_state_dir(plan.slug)
 
@@ -96,25 +129,6 @@ def launch(
         )
 
     try:
-        # Step 1: agent image. Use a committed snapshot when one exists
-        # and is present in the local daemon; otherwise build from the
-        # Dockerfile. (The gateway image is built by the orchestrator.)
-        committed = read_committed_image(plan.slug)
-        if committed and docker_mod.image_exists(committed):
-            info(f"using committed image {committed!r}")
-            plan = dataclasses.replace(
-                plan,
-                agent_provision=dataclasses.replace(plan.agent_provision, image=committed),
-            )
-        else:
-            docker_mod.build_image(
-                plan.image, _REPO_DIR,
-                dockerfile=plan.dockerfile_path,
-            )
-            docker_mod.verify_agent_image(
-                plan.image, runtime_for(plan.agent_provider_template).smoke_test,
-            )
-
         # Step 2: mint the git-gate dynamic (gitea) deploy keys, if any, before
         # provisioning the bottle's repos into the shared gateway.
         git_gate_plan = plan.git_gate_plan
@@ -207,3 +221,17 @@ def launch(
         yield bottle
     finally:
         teardown()
+
+
+def stale_checks(plan: DockerBottlePlan) -> None:
+    """Raise StaleImageError if a cached image is older than the configured
+    threshold. Only runs when image_policy is 'cached'. Called by the backend
+    class's _image_stale_checks before _launch_impl starts any resources."""
+    if plan.spec.image_policy != "cached":
+        return
+    committed = read_committed_image(plan.slug)
+    if committed and docker_mod.image_exists(committed):
+        check_stale(f"agent image {committed!r}", docker_mod.image_created_at(committed))
+        return
+    if docker_mod.image_exists(plan.image):
+        check_stale(f"agent image {plan.image!r}", docker_mod.image_created_at(plan.image))

bot_bottle/backend/docker/util.py

@@ -5,6 +5,7 @@ existence, and building images."""
 from __future__ import annotations
 
 import os
+from datetime import datetime, timezone
 import re
 import shutil
 import subprocess
@@ -197,3 +198,42 @@ def commit_container(container_name: str, image_tag: str) -> None:
             f"{(result.stderr or '').strip() or '<no stderr>'}"
         )
     info(f"committed {container_name!r} → {image_tag!r}")
+
+
+def image_created_at(ref: str) -> datetime:
+    """Return Docker's image Created timestamp as an aware UTC datetime."""
+    r = subprocess.run(
+        ["docker", "image", "inspect", "--format", "{{.Created}}", ref],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    if r.returncode != 0:
+        die(
+            f"docker image inspect for {ref!r} failed: "
+            f"{(r.stderr or '').strip() or '<no stderr>'}"
+        )
+    raw = r.stdout.strip()
+    try:
+        return _parse_docker_timestamp(raw)
+    except ValueError:
+        die(f"docker image inspect for {ref!r} returned invalid Created timestamp: {raw!r}")
+
+
+def _parse_docker_timestamp(raw: str) -> datetime:
+    text = raw.strip()
+    if text.endswith("Z"):
+        text = text[:-1] + "+00:00"
+    dot = text.find(".")
+    if dot != -1:
+        tz_plus = text.find("+", dot)
+        tz_minus = text.find("-", dot)
+        tz_candidates = [pos for pos in (tz_plus, tz_minus) if pos != -1]
+        if tz_candidates:
+            tz_pos = min(tz_candidates)
+            frac = text[dot + 1:tz_pos]
+            text = text[:dot + 1] + frac[:6].ljust(6, "0") + text[tz_pos:]
+    dt = datetime.fromisoformat(text)
+    if dt.tzinfo is None:
+        dt = dt.replace(tzinfo=timezone.utc)
+    return dt.astimezone(timezone.utc)

bot_bottle/backend/firecracker/backend.py

@@ -18,7 +18,7 @@ from ...env import ResolvedEnv
 from ...git_gate import GitGatePlan
 from ...manifest import Manifest
 from ...supervise import SupervisePlan
-from .. import ActiveAgent, BottleBackend, BottleSpec
+from .. import ActiveAgent, BottleBackend, BottleImages, BottleSpec
 from . import cleanup as _cleanup
 from . import enumerate as _enumerate
 from . import launch as _launch
@@ -92,11 +92,18 @@ class FirecrackerBottleBackend(
             stage_dir=stage_dir,
         )
 
+    def _build_or_load_images(self, plan: FirecrackerBottlePlan) -> BottleImages:
+        return BottleImages(agent=_launch.build_or_load_agent_base(plan))
+
+    def prelaunch_checks(self, plan: FirecrackerBottlePlan) -> None:
+        _launch.stale_checks(plan)
+
     @contextmanager
-    def launch(
-        self, plan: FirecrackerBottlePlan
+    def _launch_impl(
+        self, plan: FirecrackerBottlePlan, images: BottleImages,
     ) -> Generator[FirecrackerBottle, None, None]:
-        with _launch.launch(plan, provision=self.provision) as bottle:
+        assert isinstance(images.agent, Path)
+        with _launch.launch(plan, images.agent, provision=self.provision) as bottle:
             yield bottle
 
     def prepare_cleanup(self) -> FirecrackerBottleCleanupPlan:

bot_bottle/backend/firecracker/freezer.py

@@ -5,7 +5,7 @@ backend — we stream the guest root filesystem out over the control
 channel (SSH here). Unlike the other backends this needs no Docker: the
 tar *is* the resumable artifact. `resume` extracts it and rebuilds a
 fresh per-bottle ext4 with `mke2fs -d` (see `util.build_committed_rootfs_dir`
-and `launch._build_agent_base`). The bottle keeps running after the
+and `launch.build_or_load_agent_base`). The bottle keeps running after the
 snapshot.
 """
 

bot_bottle/backend/firecracker/image_builder.py

@@ -45,6 +45,13 @@ def _dockerfile_hash(dockerfile: Path) -> str:
     return hashlib.sha256(dockerfile.read_bytes()).hexdigest()[:16]
 
 
+def cached_agent_rootfs_dir(dockerfile: Path) -> Path | None:
+    """Return the ready cached rootfs for ``dockerfile``, if one exists."""
+    digest = _dockerfile_hash(dockerfile)
+    base = util.cache_dir() / "rootfs" / f"agent-{digest}"
+    return base if (base / ".bb-ready").is_file() else None
+
+
 def build_agent_rootfs_dir(
     dockerfile: Path, *, image_tag: str, smoke_test: tuple[str, ...] = (),
 ) -> Path:
@@ -58,7 +65,7 @@ def build_agent_rootfs_dir(
     silent-failure image at build time rather than at first agent use."""
     digest = _dockerfile_hash(dockerfile)
     base = util.cache_dir() / "rootfs" / f"agent-{digest}"
-    if (base / ".bb-ready").is_file():
+    if cached_agent_rootfs_dir(dockerfile) is not None:
         info(f"using cached agent rootfs {base.name}")
         return base
 

bot_bottle/backend/firecracker/launch.py

@@ -45,7 +45,8 @@ from ...git_gate import (
     provision_git_gate_dynamic_keys,
     revoke_git_gate_provisioned_keys,
 )
-from ...log import info, warn
+from ...image_cache import check_stale_path
+from ...log import die, info, warn
 from ...supervise import SUPERVISE_PORT
 from ..docker.egress import EGRESS_PORT
 from ..util import AGENT_CA_BUNDLE, AGENT_CA_PATH
@@ -64,6 +65,7 @@ _GIT_HTTP_PORT = 9420
 @contextmanager
 def launch(
     plan: FirecrackerBottlePlan,
+    agent_base: Path,
     *,
     provision: Callable[[FirecrackerBottlePlan, "FirecrackerBottle"], str | None],
 ) -> Generator[FirecrackerBottle, None, None]:
@@ -85,11 +87,9 @@ def launch(
             raise teardown_exc
 
     try:
-        # Step 1: agent rootfs. Built from the Dockerfile inside a Firecracker
-        # builder VM (buildah, no host docker); a committed snapshot is reused
-        # when present. Returns the base dir the per-bottle ext4 is made from.
-        plan, agent_base = _build_agent_base(plan)
-
+        # Step 1 (rootfs resolution/build) runs in BottleBackend.launch before
+        # this context starts resources. ``agent_base`` is the selected cache,
+        # fresh build, or committed snapshot.
         # Step 2: mint the git-gate dynamic (gitea) deploy keys, if any.
         git_gate_plan = plan.git_gate_plan
         if git_gate_plan.upstreams:
@@ -206,9 +206,7 @@ def launch(
         teardown()
 
 
-def _build_agent_base(
-    plan: FirecrackerBottlePlan,
-) -> tuple[FirecrackerBottlePlan, Path]:
+def build_or_load_agent_base(plan: FirecrackerBottlePlan) -> Path:
     """Produce the agent's base rootfs dir. Primary path: build the Dockerfile
     inside a Firecracker builder VM (buildah, no host docker), smoke-testing
     the image before export. A committed snapshot (freeze/migrate) is resumed
@@ -217,13 +215,36 @@ def _build_agent_base(
     committed_tar = committed_rootfs_path(plan.slug)
     if committed and committed_tar.is_file():
         info(f"resuming from committed rootfs {committed_tar}")
-        return plan, util.build_committed_rootfs_dir(committed_tar)
-    base = image_builder.build_agent_rootfs_dir(
-        Path(plan.dockerfile_path),
+        return util.build_committed_rootfs_dir(committed_tar)
+    dockerfile = Path(plan.dockerfile_path)
+    if plan.spec.image_policy == "cached":
+        cached = image_builder.cached_agent_rootfs_dir(dockerfile)
+        if cached is None:
+            die(
+                f"cached agent rootfs for {plan.image!r} not found; "
+                "run without --cached-images to build it"
+            )
+        info(f"using cached agent rootfs {cached.name}")
+        return cached
+    return image_builder.build_agent_rootfs_dir(
+        dockerfile,
         image_tag=plan.image,
         smoke_test=runtime_for(plan.agent_provider_template).smoke_test,
     )
-    return plan, base
+
+
+def stale_checks(plan: FirecrackerBottlePlan) -> None:
+    """Raise when the cached rootfs selected by this plan is stale."""
+    if plan.spec.image_policy != "cached":
+        return
+    committed = read_committed_image(plan.slug)
+    committed_tar = committed_rootfs_path(plan.slug)
+    if committed and committed_tar.is_file():
+        check_stale_path(f"agent rootfs {committed_tar}", committed_tar)
+        return
+    cached = image_builder.cached_agent_rootfs_dir(Path(plan.dockerfile_path))
+    if cached is not None:
+        check_stale_path(f"agent rootfs {cached}", cached / ".bb-ready")
 
 
 # --- agent guest env -------------------------------------------------

bot_bottle/backend/macos_container/backend.py

@@ -12,7 +12,7 @@ from ...env import ResolvedEnv
 from ...git_gate import GitGatePlan
 from ...supervise import SupervisePlan
 from ...manifest import Manifest
-from .. import ActiveAgent, BottleBackend, BottleSpec
+from .. import ActiveAgent, BottleBackend, BottleImages, BottleSpec
 from . import cleanup as _cleanup
 from . import enumerate as _enumerate
 from . import launch as _launch
@@ -82,11 +82,17 @@ class MacosContainerBottleBackend(
             stage_dir=stage_dir,
         )
 
+    def prelaunch_checks(self, plan: MacosContainerBottlePlan) -> None:
+        _launch.stale_checks(plan)
+
+    def _build_or_load_images(self, plan: MacosContainerBottlePlan) -> BottleImages:
+        return _launch.build_or_load_images(plan)
+
     @contextmanager
-    def launch(
-        self, plan: MacosContainerBottlePlan
+    def _launch_impl(
+        self, plan: MacosContainerBottlePlan, images: BottleImages
     ) -> Generator[MacosContainerBottle, None, None]:
-        with _launch.launch(plan, provision=self.provision) as bottle:
+        with _launch.launch(plan, images, provision=self.provision) as bottle:
             yield bottle
 
     def ensure_orchestrator(self) -> str:

bot_bottle/backend/macos_container/launch.py

@@ -53,7 +53,9 @@ from ...git_gate import (
     revoke_git_gate_provisioned_keys,
 )
 from ...git_http_backend import DEFAULT_PORT as _GIT_HTTP_PORT
+from ...image_cache import check_stale
 from ...log import die, info, warn
+from .. import BottleImages
 from ...supervise import SUPERVISE_PORT
 from ..docker.egress import EGRESS_PORT
 from ..util import AGENT_CA_BUNDLE, AGENT_CA_PATH
@@ -71,18 +73,43 @@ _REPO_DIR = str(Path(__file__).resolve().parent.parent.parent.parent)
 _AGENT_SLEEP_SECONDS = "2147483647"
 
 
+def build_or_load_images(plan: MacosContainerBottlePlan) -> BottleImages:
+    """Resolve the agent image ref for this plan. The gateway's own image is
+    built by `ensure_gateway` — it belongs to the shared singleton."""
+    committed = read_committed_image(plan.slug)
+    if committed and container_mod.image_exists(committed):
+        info(f"using committed image {committed!r}")
+        return BottleImages(agent=committed)
+    if plan.spec.image_policy == "cached":
+        if not container_mod.image_exists(plan.image):
+            die(
+                f"cached agent image {plan.image!r} not found; "
+                "run without --cached-images to build it"
+            )
+        info(f"using cached agent image {plan.image!r}")
+        return BottleImages(agent=plan.image)
+    container_mod.build_image(plan.image, _REPO_DIR, dockerfile=plan.dockerfile_path)
+    return BottleImages(agent=plan.image)
+
+
 @contextmanager
 def launch(
     plan: MacosContainerBottlePlan,
+    images: BottleImages,
     *,
     provision: Callable[[MacosContainerBottlePlan, "MacosContainerBottle"], str | None],
 ) -> Generator[MacosContainerBottle, None, None]:
-    """Build, run, register, provision, and yield an Apple Container bottle on
-    the shared per-host gateway."""
+    """Run, register, provision, and yield an Apple Container bottle on the
+    shared per-host gateway."""
     stack = ExitStack()
     bottle_for_revoke = plan.manifest.bottle
     git_gate_dir_for_revoke = git_gate_state_dir(plan.slug)
 
+    plan = dataclasses.replace(
+        plan,
+        agent_provision=dataclasses.replace(plan.agent_provision, image=str(images.agent)),
+    )
+
     def teardown() -> None:
         teardown_exc: BaseException | None = None
         try:
@@ -95,8 +122,6 @@ def launch(
             raise teardown_exc
 
     try:
-        plan = _build_images(plan)
-
         # Step 1: the per-host singletons. Must precede the agent run — its
         # proxy env needs the gateway's address at `container run` time.
         endpoint = ensure_gateway()
@@ -177,22 +202,19 @@ def launch(
         teardown()
 
 
-def _build_images(plan: MacosContainerBottlePlan) -> MacosContainerBottlePlan:
-    """Build the agent image. The gateway's own image is built by
-    `ensure_gateway` — it belongs to the shared singleton, not to a bottle."""
+
+def stale_checks(plan: MacosContainerBottlePlan) -> None:
+    """Raise StaleImageError if a cached image is older than the configured
+    threshold. Only runs when image_policy is 'cached'. Called by the backend
+    class's _image_stale_checks before _launch_impl starts any resources."""
+    if plan.spec.image_policy != "cached":
+        return
     committed = read_committed_image(plan.slug)
     if committed and container_mod.image_exists(committed):
-        info(f"using committed image {committed!r}")
-        return dataclasses.replace(
-            plan,
-            agent_provision=dataclasses.replace(
-                plan.agent_provision, image=committed,
-            ),
-        )
-    container_mod.build_image(
-        plan.image, _REPO_DIR, dockerfile=plan.dockerfile_path,
-    )
-    return plan
+        check_stale(f"agent image {committed!r}", container_mod.image_created_at(committed))
+        return
+    if container_mod.image_exists(plan.image):
+        check_stale(f"agent image {plan.image!r}", container_mod.image_created_at(plan.image))
 
 
 def _provision_git_gate_keys(

bot_bottle/backend/macos_container/util.py

@@ -10,6 +10,7 @@ import shutil
 import subprocess
 import tempfile
 import time
+from datetime import datetime, timezone
 from typing import Iterable
 
 from ...log import die, info
@@ -611,6 +612,39 @@ def image_id(ref: str) -> str:
     raise AssertionError("unreachable")
 
 
+def image_created_at(ref: str) -> datetime:
+    """Return the image creation timestamp as an aware UTC datetime.
+
+    Parses the `created` field from `container image inspect` JSON output."""
+    result = subprocess.run(
+        [_CONTAINER, "image", "inspect", ref],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    if result.returncode != 0:
+        die(
+            f"container image inspect for {ref!r} failed: "
+            f"{(result.stderr or '').strip() or '<no stderr>'}"
+        )
+    try:
+        data = json.loads(result.stdout or "{}")
+    except json.JSONDecodeError as exc:
+        die(f"container image inspect for {ref!r} returned malformed JSON: {exc}")
+    if isinstance(data, list) and data:
+        data = data[0]
+    if isinstance(data, dict):
+        value = data.get("created") or data.get("Created")
+        if isinstance(value, str) and value:
+            try:
+                ts = value.rstrip("Z")
+                return datetime.fromisoformat(ts).replace(tzinfo=timezone.utc)
+            except ValueError:
+                pass
+    die(f"container image inspect for {ref!r} did not include a creation timestamp")
+    raise AssertionError("unreachable")
+
+
 def save(ref: str, output: str) -> None:
     subprocess.run([_CONTAINER, "image", "save", ref, "-o", output], check=True)
 

bot_bottle/cli/__init__.py

@@ -38,6 +38,13 @@ COMMANDS = {
     "supervise": cmd_supervise,
 }
 
+# Commands that manage host prerequisites (or are otherwise store-free) and
+# must run before — or without — a migrated DB. `backend` provisions/probes
+# the host (TAP pool, /dev/kvm, firecracker) and never opens the store, so
+# gating it on the schema breaks preflight on a fresh CI runner where stdin
+# isn't a TTY and the migration prompt can't be answered.
+NO_MIGRATION_COMMANDS = frozenset({"backend"})
+
 
 def usage() -> None:
     sys.stderr.write(f"usage: {PROG} <command> [args...]\n\n")
@@ -80,7 +87,7 @@ def main(argv: list[str] | None = None) -> int:
         usage()
         die(f"unknown command: {command}")
     mgr = StoreManager.instance()
-    if not mgr.is_migrated():
+    if command not in NO_MIGRATION_COMMANDS and not mgr.is_migrated():
         sys.stderr.write("bot-bottle: database schema is out of date\n")
         sys.stderr.write("Migrate now? [y/N] ")
         sys.stderr.flush()

bot_bottle/cli/start.py

@@ -36,6 +36,7 @@ from ..bottle_state import (
     is_preserved,
     mark_preserved,
 )
+from ..image_cache import StaleImageError
 from ..log import info, die
 from ..manifest import Manifest, ManifestIndex
 from ._common import PROG, USER_CWD, read_tty_line
@@ -74,6 +75,14 @@ def cmd_start(argv: list[str]) -> int:
             "skip all prompts. For orchestrators, CI, and webhooks."
         ),
     )
+    parser.add_argument(
+        "--cached-images",
+        action="store_true",
+        help=(
+            "quickstart with existing local agent and sidecar images; "
+            "only valid with --headless"
+        ),
+    )
     parser.add_argument(
         "--bottle",
         action="append",
@@ -106,6 +115,8 @@ def cmd_start(argv: list[str]) -> int:
         help="agent name defined in bot-bottle.json (omit to pick interactively)",
     )
     args = parser.parse_args(argv)
+    if args.cached_images and not args.headless:
+        die("--cached-images is only supported with --headless")
 
     dry_run = args.dry_run or os.environ.get("BOT_BOTTLE_DRY_RUN") == "1"
     if args.no_cache or os.environ.get("BOT_BOTTLE_NO_CACHE") == "1":
@@ -158,6 +169,10 @@ def cmd_start(argv: list[str]) -> int:
     label, color = tui.name_color_modal(default_label=agent_name)
     label, color = _resolve_unique_label(label, color)
 
+    image_policy = _select_image_policy()
+    if image_policy is None:
+        return 0
+
     spec = BottleSpec(
         manifest=manifest,
         agent_name=agent_name,
@@ -166,6 +181,7 @@ def cmd_start(argv: list[str]) -> int:
         label=label,
         color=color,
         bottle_names=bottle_names,
+        image_policy=image_policy,
     )
     return _launch_bottle(
         spec,
@@ -226,6 +242,7 @@ def _start_headless(
         color=args.color or "",
         bottle_names=bottle_names,
         headless=True,
+        image_policy="cached" if args.cached_images else "fresh",
     )
     return _launch_bottle(
         spec,
@@ -406,6 +423,13 @@ def _text_prompt_yes() -> bool:
     return reply in ("y", "Y", "yes", "YES")
 
 
+def _select_image_policy() -> str | None:
+    return tui.filter_select(
+        ["fresh", "cached"],
+        title="Select image startup mode",
+    )
+
+
 def _text_render_preflight():
     def _render(plan: DockerBottlePlan, backend_name: str) -> None:
         print(file=sys.stderr)
@@ -548,6 +572,15 @@ def _launch_bottle(
             return 0
 
         backend = get_bottle_backend(backend_name)
+        try:
+            backend.prelaunch_checks(plan)
+        except StaleImageError as exc:
+            if assume_yes:
+                die(str(exc))
+            sys.stderr.write(f"bot-bottle: {exc}\nLaunch anyway? [y/N] ")
+            sys.stderr.flush()
+            if read_tty_line() not in ("y", "Y", "yes", "YES"):
+                return 0
         with backend.launch(plan) as bottle:
             agent_provider_template = getattr(plan, "agent_provider_template", "claude")
             extra_args: tuple[str, ...] = ()
@@ -566,10 +599,6 @@ def _launch_bottle(
                 f"session ended (exit {exit_code}); "
                 f"container {bottle.name} will be removed"
             )
-            # While the container is still alive: always snapshot the
-            # transcript and — if the agent exited non-zero — mark
-            # the state for preservation. This picks up crashes /
-            # Ctrl-Cs / OOM kills before cleanup removes the state dir.
             if agent_provider_template == "claude":
                 capture_claude_session_state(identity, exit_code)
         return 0

bot_bottle/config_store.py

@@ -0,0 +1,71 @@
+"""SQLite-backed bot-bottle configuration store."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+try:
+    from .db_store import DbStore
+    from .migrations import TableMigrations
+    from .paths import host_db_path
+except ImportError:
+    from db_store import DbStore  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
+    from migrations import TableMigrations  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
+    from paths import host_db_path  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
+
+
+DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS = 1
+
+
+class ConfigStore(DbStore):
+    """SQLite configuration for host-side bot-bottle settings."""
+
+    def __init__(self, db_path: Path | None = None) -> None:
+        migrations = TableMigrations("config_store", [
+            # v1 — host-side bot-bottle settings
+            """
+            CREATE TABLE IF NOT EXISTS bot_bottle_config (
+                id INTEGER PRIMARY KEY CHECK (id = 1),
+                cached_image_stale_warning_days INTEGER NOT NULL DEFAULT 1
+            )
+            """,
+        ])
+        super().__init__(db_path or host_db_path(), migrations)
+
+    def cached_image_stale_warning_days(self) -> int:
+        if not self.db_path.is_file():
+            return DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS
+        with self._connect() as conn:
+            row = conn.execute(
+                """
+                SELECT cached_image_stale_warning_days
+                FROM bot_bottle_config
+                WHERE id = 1
+                """,
+            ).fetchone()
+        if row is None:
+            return DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS
+        try:
+            return int(row["cached_image_stale_warning_days"])
+        except (TypeError, ValueError):
+            return DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS
+
+    def set_cached_image_stale_warning_days(self, days: int) -> Path:
+        with self._connect() as conn:
+            conn.execute(
+                """
+                INSERT INTO bot_bottle_config (id, cached_image_stale_warning_days)
+                VALUES (1, ?)
+                ON CONFLICT(id) DO UPDATE SET
+                    cached_image_stale_warning_days = excluded.cached_image_stale_warning_days
+                """,
+                (days,),
+            )
+        self._chmod()
+        return self.db_path
+
+
+__all__ = [
+    "DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS",
+    "ConfigStore",
+]

bot_bottle/git_gate_render.py

@@ -419,18 +419,24 @@ PY
 while IFS=' ' read -r old new ref; do
   [ -z "$ref" ] && continue
   [ "$new" = "$zero" ] && continue
-  if [ "$old" = "$zero" ]; then
-    # New ref: scan only the commits this push introduces — those
-    # reachable from $new but not from any ref the gate already has.
-    # Everything already on the gate arrived via upstream mirror-fetch
-    # or a previously gitleaks-scanned push, so it's already-upstream
-    # or already-scanned; re-scanning it (the old `$new` full-ancestry
-    # range) only resurfaces historical findings and blocks every new
-    # branch. See PRD 0028 / issue #106.
-    log_opts="$new --not --all"
-  else
-    log_opts="$old..$new"
-  fi
+  # Scan only the commits this push introduces — those reachable from
+  # $new but not from any ref the gate already has. Everything already
+  # on the gate arrived via upstream mirror-fetch or a previously
+  # gitleaks-scanned push, so it's already-upstream or already-scanned;
+  # re-scanning it only resurfaces historical fixture findings.
+  #
+  # Applies to both new refs and updates. The old existing-branch range
+  # `$old..$new` walks commits reachable from the new tip but not the
+  # *old branch tip*: on a rebase/force-push onto a freshly-advanced
+  # main that pulls in all of main's new history (incl. the deliberate
+  # sandbox-escape gitleaks fixtures), blocking the push. `--not --all`
+  # excludes anything already on the gate regardless of ancestry, so it
+  # is also correct for non-fast-forward pushes (a rebase can skip
+  # commits off the direct path). Security-equivalent per PRD 0028's
+  # analysis: the bare repo's refs come only from trusted upstream
+  # mirror-fetch or gitleaks-gated pushes.
+  # See PRD 0028 (open question) / issues #106, #346.
+  log_opts="$new --not --all"
   echo "git-gate: gitleaks scanning $ref ($log_opts)" >&2
   if ! gitleaks git --log-opts="$log_opts" --no-banner --redact 1>&2; then
     echo "git-gate: gitleaks rejected push to $ref" >&2

bot_bottle/image_cache.py

@@ -0,0 +1,42 @@
+"""Shared helpers for cached-image quickstart stale checks."""
+
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from pathlib import Path
+
+try:
+    from .config_store import ConfigStore
+except ImportError:
+    from config_store import ConfigStore  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
+
+
+class StaleImageError(Exception):
+    """Raised when a cached image or artifact exceeds the configured staleness
+    threshold. Callers can catch this to prompt interactively; headless paths
+    let it propagate as a fatal error."""
+
+
+def check_stale(label: str, created_at: datetime) -> None:
+    """Raise StaleImageError if `created_at` is older than the configured
+    stale-warning threshold. Negative threshold disables the check."""
+    threshold_days = ConfigStore().cached_image_stale_warning_days()
+    if threshold_days < 0:
+        return
+    now = datetime.now(timezone.utc)
+    created = created_at.astimezone(timezone.utc)
+    age = now - created
+    if age.total_seconds() <= threshold_days * 86400:
+        return
+    raise StaleImageError(
+        f"cached {label} is {age.days} day(s) old; "
+        "quickstart does not verify it matches the current Dockerfile/context"
+    )
+
+
+def check_stale_path(label: str, path: Path) -> None:
+    """Raise StaleImageError if `path`'s mtime exceeds the staleness threshold."""
+    check_stale(label, datetime.fromtimestamp(path.stat().st_mtime, tz=timezone.utc))
+
+
+__all__ = ["StaleImageError", "check_stale", "check_stale_path"]

bot_bottle/store_manager.py

@@ -6,9 +6,11 @@ from pathlib import Path
 
 try:
     from .audit_store import AuditStore
+    from .config_store import ConfigStore
     from .queue_store import QueueStore
 except ImportError:
     from audit_store import AuditStore  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
+    from config_store import ConfigStore  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
     from queue_store import QueueStore  # type: ignore[import-not-found]  # pylint: disable=import-error,no-name-in-module
 
 _instance: StoreManager | None = None
@@ -47,11 +49,13 @@ class StoreManager:
         return (
             QueueStore("", self.db_path).is_migrated()
             and AuditStore(self.db_path).is_migrated()
+            and ConfigStore(self.db_path).is_migrated()
         )
 
     def migrate(self) -> None:
         QueueStore("", self.db_path).migrate()
         AuditStore(self.db_path).migrate()
+        ConfigStore(self.db_path).migrate()
 
 
 __all__ = ["StoreManager"]

docs/decisions/0005-issues-own-tracker-metadata.md

@@ -0,0 +1,57 @@
+# ADR 0005: Keep tracker metadata on issues
+
+- **Status:** Accepted
+- **Date:** 2026-07-18
+- **Deciders:** didericis
+
+## Context
+
+Gitea exposes labels on both issues and pull requests. Applying the same labels
+to both copies planning metadata, creates a synchronization obligation, and
+makes disagreements between the two records possible. At the same time,
+unlabelled objects look accidental unless the repository states which object
+owns the metadata.
+
+The repository already uses issues as work items and PRs as implementations of
+those work items. At this decision's cutoff, all open PRs reference issues, but
+121 of 219 historically merged PRs do not. Manufacturing retrospective issues
+for that history would create records that never participated in planning and
+would make the issue history less truthful.
+
+## Decision
+
+Issues are the canonical tracker records and own labels. Every issue has at
+least one label. An issue opened or left without labels receives
+`Status/Needs Triage` automatically until it is classified.
+
+Pull requests carry no labels. Every new PR deliberately references at least
+one existing issue in its title or description with one of these forms:
+
+- `Closes #123`, `Fixes #123`, or `Resolves #123` when merging completes it.
+- `Part of #123`, `Related to #123`, `Refs #123`, or `References #123` when it
+  contributes without completing it.
+
+Gitea Actions enforces both PR rules as a status check and repairs the empty
+issue-label state. Branch protection makes the PR policy check required.
+
+The policy applies from 2026-07-18 onward. Existing issues may be labelled as
+they are encountered, but closed PRs are grandfathered: no retrospective
+issues or PR labels are created solely to make history conform.
+
+## Consequences
+
+- Classification, priority, and workflow metadata have one source of truth.
+- A PR's issue link is the navigation path to its planning metadata.
+- Multi-PR issues do not require copied or synchronized labels.
+- `Status/Needs Triage` is an intentional fallback, not a final
+  classification.
+- Direct issue creation remains convenient; automation repairs a missing label
+  immediately after creation because Gitea has no native required-label rule.
+- The required check must be configured in branch protection after this
+  workflow lands.
+
+## Links
+
+- Issue #405.
+- `.gitea/workflows/tracker-policy.yml`.
+- `scripts/tracker_policy.py`.

docs/research/agent-sandbox-landscape.md

@@ -6,27 +6,61 @@ general AI-agent sandbox / containment projects — some Claude-specific,
 some agent-agnostic, some hosted SaaS — and contrasts them with
 bot-bottle's design.
 
-Research conducted 2026-05-11.
+Research conducted 2026-05-11. CubeSandbox added 2026-07-18 (see its
+per-project note and the addendum at the end). Also updated 2026-07-18:
+bot-bottle no longer uses **pipelock** — outbound DLP is now bot-bottle's
+own (deliberately simple) egress scanner (a mitmproxy addon with custom
+detectors, PRD 0017 / 0053), and git-push secret scanning is handled by
+**gitleaks** in the git-gate. "pipelock" below has been replaced with the
+current mechanism; it survives only in older PRDs as history.
+
+Updated again 2026-07-18: six additional tools added (Cleanroom,
+container-use, Docker sbx, Anthropic srt, Microsoft AGT, Open Agent
+Passport); an **Agent-tailored policy** row added to the comparison table;
+a separate Governance layers section added for AGT and OAP. See the
+second addendum at the end.
 
 ## Summary
 
-Eight projects surveyed. None duplicate bot-bottle's combination of
-local Docker, declarative JSON manifest, per-agent egress allowlist via
-pipelock, and bottle/agent split. Two clusters stand out:
+Fifteen projects surveyed across two categories: isolation/sandbox tools
+and governance/pre-action authorization layers (the latter don't provide
+VM or container isolation but do per-agent policy enforcement at the
+tool-call level). None duplicate bot-bottle's combination of local
+VM-per-bottle isolation, a declarative per-role manifest, per-agent
+egress allowlist + outbound-content DLP, bottle/agent split, and the
+composable `extends:` policy model. Three clusters stand out:
 
 - **Closest neighbours** — agent-safehouse and litterbox: local,
   single-user, thin wrappers over an existing OS primitive
   (`sandbox-exec`, Podman + Landlock).
-- **Different category** — tilde.run (hosted SaaS), boxlite and
-  microsandbox (microVM libraries for platform builders), endo-familiar
+- **Different category (isolation)** — tilde.run (hosted SaaS), boxlite
+  and microsandbox (microVM libraries for platform builders), CubeSandbox
+  (self-hosted multi-tenant microVM service), endo-familiar
   (capability-security paradigm, no OS isolation).
+- **New: governance/pre-action layers** — Microsoft AGT and Open Agent
+  Passport (OAP): framework-embedded tool-call interceptors with
+  per-agent declarative policy. Closest competitors on agent-tailored
+  policy, but operate at the tool-call level rather than providing
+  network/filesystem isolation; they complement rather than substitute.
 
-The microVM cluster (matchlock, smolmachines, boxlite, microsandbox) is
-the most relevant for the v2 isolation discussion in
+The microVM cluster (matchlock, smolmachines, boxlite, microsandbox,
+CubeSandbox) is the most relevant for the v2 isolation discussion in
 [`stronger-isolation-alternatives.md`](stronger-isolation-alternatives.md):
 libkrun and Apple's Virtualization.framework have made local microVMs
-ergonomic enough that a `"runtime": "microvm"` option on a bottle is now
-plausible without a heavy stack.
+ergonomic enough that microVMs are **now bot-bottle's default backend**
+(Firecracker on KVM Linux, Apple Container on macOS), with Docker kept
+only as a legacy fallback for CI / hosts without KVM or Apple Container.
+That discussion has since shipped, not just been theorized.
+
+**The one that matters most for positioning is CubeSandbox** — it is the
+first surveyed project to ship bot-bottle's would-be wedge (default-deny
+egress allowlist + full audit logs + in-flight credential custody so keys
+never enter the sandbox) *combined with* per-sandbox microVM isolation,
+open-source under Apache 2.0, with Tencent Cloud behind it and 10.4k
+stars. It's a self-hosted multi-tenant service for platform builders, not
+a single-user declarative tool, so it doesn't collide head-on — but it
+narrows the "nobody else bundles egress custody + credential injection"
+claim that the monetization positioning leans on. See the addendum.
 
 ## Per-project notes
 
@@ -155,67 +189,272 @@ plausible without a heavy stack.
   also supported.
 - **Maturity**: Active through April 2026.
 
+### CubeSandbox *(added 2026-07-18)*
+- **Source**: https://github.com/TencentCloud/CubeSandbox ;
+  HN launch https://news.ycombinator.com/item?id=47863430
+- **License**: Apache 2.0 (~10.4k stars). By Tencent Cloud; described as
+  "battle-tested, production-ready" infra already running in Tencent
+  Cloud. Rust / Go / C.
+- **Isolation**: MicroVMs via RustVMM + KVM — "each sandbox gets its own
+  Guest OS kernel, no Docker shared-kernel escapes." Hardware-level
+  isolation, dedicated kernel per instance.
+- **Locality**: Self-hosted, but **server/cluster-oriented**, not a
+  single-user local CLI. Deploy guides target PVM cloud VMs, bare metal,
+  and dev. A single 96-vCPU host is claimed to run 2,000+ concurrent
+  sandboxes.
+- **Agent integration**: **Drop-in E2B SDK replacement** (single env-var
+  change) — the headline compatibility claim. OpenClaw assistant
+  integration; general LLM-code execution. Aimed at platform builders,
+  not one developer's laptop.
+- **Config**: Programmatic via the E2B-compatible SDK. No declarative
+  manifest.
+- **Network policy**: This is the striking part — **domain allowlists,
+  instant block on unauthorized egress, full audit logs, per-sandbox
+  traffic tokens, policy-routing egress**, enforced by an eBPF-based
+  virtual switch giving kernel-level network isolation. Closest match yet
+  to bot-bottle's own default-deny + per-bottle allowlist egress model.
+- **Credentials**: **Credential vault** — agents call external APIs / LLMs
+  while "keys never enter the sandbox, model context, or logs." Same
+  in-flight-injection idea as matchlock, but productized as a vault.
+- **Performance**: <60ms cold start (claimed 2.5–50× faster than
+  alternatives), <5MB memory per instance; millisecond snapshot rollback
+  is upcoming.
+- **Maturity**: Open-sourced July 2026 off production Tencent Cloud use;
+  most-starred project in this set (~10.4k).
+
+### Cleanroom *(added 2026-07-18)*
+- **Source**: https://github.com/buildkite/cleanroom
+- **License**: Apache 2.0
+- **Isolation**: MicroVM — Firecracker on Linux, Virtualization.framework
+  on macOS. Digest-pinned OCI images.
+- **Locality**: Self-hosted server (CI-oriented).
+- **Agent integration**: Generic process sandbox; CI-first, not a
+  Claude/agent wrapper.
+- **Config**: `cleanroom.yaml` in the repo being sandboxed defines egress
+  rules, resources, and network policy. Cleanroom resolves this from the
+  commit being run.
+- **Network policy**: Default-deny + per-repo hostname allowlist (resolved
+  from DNS answers + destination IP:port). Co-hosted services on the same
+  IP:port are not distinguished. OIDC-backed auth for remote servers.
+- **Credentials**: Host-side only; not injected in-flight but not present
+  in the VM.
+- **Notable**: Policy lives in the *repo being sandboxed*, not in an
+  agent-role definition — closer to per-repo scoping than per-role.
+  Supports Docker-inside-sandbox (`services.docker.required: true`), OIDC
+  authorization, suspend/resume lifecycle.
+- **Maturity**: Active Buildkite product.
+
+### container-use *(added 2026-07-18)*
+- **Source**: https://github.com/dagger/container-use
+- **License**: Apache 2.0
+- **Isolation**: Docker container per agent + git worktree per agent.
+  Containers share the host kernel; stronger than bare host but weaker
+  than microVM.
+- **Locality**: Local.
+- **Agent integration**: MCP stdio server — Claude Code, Cursor, Windsurf.
+  `claude mcp add container-use -- container-use stdio`.
+- **Config**: None for security policy. Environments are provisioned on
+  demand; no allowlist or credential config.
+- **Network policy**: Not addressed.
+- **Notable**: Per-agent git branches (`container-use/<env_name>`);
+  parallel agents without filesystem conflict; real-time log visibility
+  and terminal attach for intervention; git-based review workflow.
+  Oriented toward parallel development safety, not security containment.
+- **Maturity**: Early development, active.
+
+### Docker sbx *(added 2026-07-18)*
+- **Source**: Docker proprietary (`sbx` CLI, separate from `docker`).
+- **License**: Proprietary.
+- **Isolation**: MicroVM (Docker's own implementation) — each session gets
+  its own kernel, Docker daemon inside the VM, and filesystem.
+- **Locality**: Local (macOS and Windows; does not require Docker Desktop).
+- **Agent integration**: Explicit wrapper — Claude Code, Codex, Gemini
+  CLI, Copilot CLI, Kiro. Launches agent inside the VM with
+  `--dangerously-skip-permissions` by default.
+- **Config**: Open / Balanced / Locked Down network presets at launch. No
+  per-role manifest.
+- **Network policy**: Default-deny; preset levels control strictness. TUI
+  dashboard shows a live log of every outbound connection (allowed and
+  blocked) with point-and-click allow/block for hosts.
+- **Credentials**: OS keychain + host-side proxy injection — API keys
+  never enter the VM.
+- **Notable**: Best DX among microVM tools (one command, works like native
+  yolo Claude but inside a VM); branch mode creates a git worktree in
+  `.sbx/`. Network policy is preset-based, not role-declarative.
+- **Maturity**: GA 2026.
+
+### Anthropic srt *(added 2026-07-18)*
+- **Source**: https://github.com/anthropic-experimental/sandbox-runtime
+  (`@anthropic-ai/sandbox-runtime` on npm, `sandbox-runtime` on PyPI)
+- **License**: Apache 2.0 (experimental).
+- **Isolation**: OS-level only — Seatbelt (`sandbox-exec`) on macOS,
+  bubblewrap on Linux, WFP (Windows Filtering Platform) account-fenced on
+  Windows. **No container or VM.** Lowest overhead in the set.
+- **Locality**: Local.
+- **Agent integration**: Claude Code's sandboxed bash tool uses this
+  internally. Can wrap any arbitrary process (`srt <command>`). Cloud
+  Claude Code sessions use full microVMs instead.
+- **Config**: Programmatic per-invocation — allow/deny path lists for
+  filesystem; allow/denylist for network (HTTP proxy + SOCKS5).
+- **Network policy**: Proxy-based filtering (HTTP + SOCKS5); domain
+  allowlist/denylist enforced at proxy layer. Custom proxy supported
+  (e.g. mitmproxy for inspection + audit). Processes that ignore proxy
+  env vars may bypass filtering on some platforms.
+- **Notable**: Cross-platform (macOS/Linux/Windows); wraps any process,
+  not just agents; no role/manifest concept. Annotated as a research
+  preview — APIs may change.
+- **Maturity**: Early research preview.
+
+## Governance / pre-action authorization layers
+
+These two tools don't provide VM or filesystem isolation; they intercept
+tool calls before execution and evaluate them against a per-agent
+declarative policy. They are the closest competitors on **agent-tailored
+policy** and complement isolation sandboxes rather than substituting for
+them.
+
+### Microsoft Agent Governance Toolkit (AGT) *(added 2026-07-18)*
+- **Source**: https://github.com/microsoft/agent-governance-toolkit
+- **License**: MIT (~3.3k stars, open-sourced April 2, 2026).
+- **Isolation**: None (OS/VM). Execution rings (0–3, inspired by CPU
+  privilege levels) control what an agent can do at the framework layer.
+  MCP security gateway treats MCP traffic as an untrusted boundary.
+- **Locality**: Embedded in the agent framework (Python, TypeScript, .NET,
+  Rust, Go; 20+ framework adapters).
+- **Agent integration**: Framework-agnostic. Plugs into Semantic Kernel,
+  AutoGen, and others as a middleware layer.
+- **Config**: YAML policy per agent — tools can be `allowed`, `denied`,
+  `sandboxed`, or routed through an `approval` step. Every action passes
+  through a governance gate checking: agent DID, trust score, risk tier,
+  requested tool, action type, and policy rules.
+- **Network policy**: Not directly — operates at tool-call level.
+- **Credentials**: Per-agent DID (Ed25519 decentralized identifier); agent
+  does not borrow a human's credentials.
+- **Notable**: Dynamic trust score (0–1,000, behavioral decay) —
+  privilege follows observed behaviour, not just provisioning. Covers all
+  10 OWASP Agentic Top 10 risks. Kill switch + SLO monitoring. Sub-ms
+  policy enforcement.
+- **Maturity**: MIT, ~3.3k ⭐, v3.7.0 May 2026.
+
+### Open Agent Passport (OAP) *(added 2026-07-18)*
+- **Source**: https://github.com/aporthq/aport-spec ; spec at
+  https://api.aport.io/spec/spec/oap/oap-spec.md/ ; arXiv 2603.20953
+- **License**: Open specification.
+- **Isolation**: None. Pre-action hook only — intercepts tool calls
+  synchronously before execution, evaluates against a cloud-registry
+  declarative policy, fails closed.
+- **Locality**: Local hook + cloud policy registry.
+- **Agent integration**: Framework-agnostic; hook pattern.
+- **Config**: Declarative policy rules in a cloud registry (evaluated in
+  order; first failing rule denies). Ed25519-signed, hash-chained audit
+  records per decision.
+- **Network policy**: Not directly.
+- **Notable**: 53ms median authorization decision (N=1,000). In an
+  adversarial testbed ($5,000 bounty, 1,151 sessions), social engineering
+  succeeded 74.6% of the time under a permissive policy; under a
+  restrictive OAP policy, 0% success across 879 attempts. Assumes
+  framework runtime is not compromised.
+- **Maturity**: Specification + reference implementation, 2026.
+
 ## Comparison table
 
-| Axis | bot-bottle | endo-familiar | litterbox | agent-safehouse | matchlock | tilde.run | boxlite | microsandbox | smolmachines |
-|---|---|---|---|---|---|---|---|---|---|
-| Isolation | Docker + internal net + pipelock; gVisor if present | Object-capability (no OS isolation) | Podman + opt. Landlock | macOS `sandbox-exec` | MicroVM (Firecracker / Virt.fw) | Hosted container (unverified) | MicroVM (KVM / Hypervisor.fw) | MicroVM (libkrun) | MicroVM (libkrun / KVM) |
-| Local vs hosted | Local | Local | Local (Linux) | Local (macOS) | Local | Hosted SaaS | Local | Local | Local |
-| Open source | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 | MIT | No | Apache 2.0 | Apache 2.0 | Apache 2.0 |
-| Agent target | Claude Code | Generic (demo) | Generic | Multi-agent wrapper | Generic (+ Claude/OpenAI SDKs) | Claude focus | Generic | Claude + Cursor (MCP/Skills) | Generic (AGENTS.md) |
-| Network policy | Default-deny via pipelock + per-bottle allowlist + DLP | Capability model only | Limited | Not addressed | Default-deny + allowlist + secret-injecting proxy | Default-deny + logging | Per-VM net (unverified) | Not documented | Off by default + allowlist |
-| Parallel agents | Yes (one bottle per agent) | n/a | Not addressed | One at a time | Multiple VMs | Yes (dashboard) | SDK-level | SDK-level | Architectural |
-| Config | JSON manifest (bottles + agents) | Programmatic refs | CLI wizard | Profile files / shell fns | CLI / SDK | DSL + CLI + SDK | SDK | CLI / SDK / MCP | TOML Smolfile |
-| Maturity | Active May 2026 | Research (2022+) | Early (~66 ⭐) | Active (~1.4k ⭐) | Experimental (~574 ⭐) | Private preview | YC, ~4.7k ⭐ | YC, ~6k ⭐, beta | ~3.1k ⭐ |
+*Isolation/sandbox tools only. AGT and OAP are governance layers — see their per-project notes above.*
+
+| Axis | bot-bottle | endo-familiar | litterbox | agent-safehouse | matchlock | tilde.run | boxlite | microsandbox | smolmachines | CubeSandbox | Cleanroom | container-use | Docker sbx | Anthropic srt |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| Isolation | MicroVM per bottle default (Firecracker/KVM on Linux, Apple Container on macOS) + own egress DLP scanner; Docker legacy fallback, gVisor there if present | Object-capability (no OS isolation) | Podman + opt. Landlock | macOS `sandbox-exec` | MicroVM (Firecracker / Virt.fw) | Hosted container (unverified) | MicroVM (KVM / Hypervisor.fw) | MicroVM (libkrun) | MicroVM (libkrun / KVM) | MicroVM (RustVMM / KVM) | MicroVM (Firecracker / Virt.fw) | Docker container + git worktree | MicroVM (proprietary) | OS-level (Seatbelt / bubblewrap / WFP) — no container |
+| Local vs hosted | Local | Local | Local (Linux) | Local (macOS) | Local | Hosted SaaS | Local | Local | Local | Self-hosted (server/cluster) | Self-hosted server | Local | Local | Local |
+| Open source | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 | MIT | No | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 | Proprietary | Apache 2.0 (experimental) |
+| Agent target | Claude Code | Generic (demo) | Generic | Multi-agent wrapper | Generic (+ Claude/OpenAI SDKs) | Claude focus | Generic | Claude + Cursor (MCP/Skills) | Generic (AGENTS.md) | E2B-compatible (platform builders) | CI / generic process | Claude Code, Cursor, Windsurf (MCP) | Claude Code, Codex, Gemini CLI, Copilot, Kiro | Claude Code (and any process) |
+| Network policy | Default-deny via own egress scanner + per-bottle allowlist + content DLP + gitleaks on git push | Capability model only | Limited | Not addressed | Default-deny + allowlist + secret-injecting proxy | Default-deny + logging | Per-VM net (unverified) | Not documented | Off by default + allowlist | Default-deny allowlist + instant egress block + audit logs + per-sandbox tokens (eBPF) + credential vault | Default-deny + per-repo host allowlist (cleanroom.yaml) | Not addressed | Default-deny; Open / Balanced / Locked Down presets; live TUI network panel | Proxy-based allowlist/denylist (HTTP + SOCKS5); custom proxy supported |
+| Parallel agents | Yes (one bottle per agent) | n/a | Not addressed | One at a time | Multiple VMs | Yes (dashboard) | SDK-level | SDK-level | Architectural | Yes (2,000+/host claimed) | Yes (server model) | Yes (per-agent containers + worktrees) | Yes | Yes |
+| Long-running posture | Persistent by default (named, supervised) | n/a (demo) | Session (up while in use) | Per-invocation | Ephemeral VM per run | Per-run (versioned) | Ephemeral + snapshot/fork | Ephemeral / on-demand | Named persistent by default | Ephemeral + auto pause/resume | Per-run + suspend/resume | Per-agent container (ephemeral) | Per-session; branch mode creates git worktree in .sbx/ | Per-invocation |
+| DX: run Claude yolo-style | One command → interactive yolo Claude (`start <agent>`, `--dangerously-skip-permissions` default) | n/a (lib demo) | Wizard + build, then run claude inside (Linux only) | One-command wrapper (`safehouse claude --dangerously-skip-permissions`) | CLI: run a cmd in a VM (not a Claude wrapper) | Hosted (`tilde exec`), not local-native | SDK code required (build the run yourself) | CLI/MCP: sandbox-as-a-tool for the agent, not a wrapper around it | SSH into a named machine, run claude there | Stand up a cluster + drive via E2B SDK | CI-oriented, not a Claude wrapper | MCP server: `claude mcp add container-use -- container-use stdio` | One command: `sbx` wraps claude with `--dangerously-skip-permissions` default | Library/wrapper, not a standalone CLI |
+| Config | JSON manifest (bottles + agents) | Programmatic refs | CLI wizard | Profile files / shell fns | CLI / SDK | DSL + CLI + SDK | SDK | CLI / SDK / MCP | TOML Smolfile | E2B-compatible SDK | cleanroom.yaml in repo | None (no policy config) | Preset levels at launch | Programmatic per-invocation (allow/deny lists) |
+| Agent-tailored policy | Yes — bottle/agent split; declarative per-role egress + credentials; composable via `extends:` | Partial — capability model scopes per-agent, but no declarative role manifest | No | Partial — per-agent profile files (Seatbelt); no egress | No | Yes — per-agent DSL RBAC (allow/deny/approve per action/repo/agent) | No | No | No | No — per-sandbox SDK config, not role-scoped | Partial — per-repo cleanroom.yaml, not per-role | No | No — network presets only | No |
+| Maturity | Active July 2026 | Research (2022+) | Early (~66 ⭐) | Active (~1.4k ⭐) | Experimental (~574 ⭐) | Private preview | YC, ~4.7k ⭐ | YC, ~6k ⭐, beta | ~3.1k ⭐ | Tencent, prod, ~10.4k ⭐ | Active (Buildkite product) | Early development | GA 2026 | Early research preview |
 
 ## What's closest, what's different
 
 **Closest in design and scope.** agent-safehouse and litterbox sit
 nearest bot-bottle: local, single-user, thin wrappers over an
 existing OS primitive, low-dep. The split is the isolation primitive —
-bot-bottle uses Docker + pipelock egress (plus gVisor where
-available); agent-safehouse uses `sandbox-exec`; litterbox uses Podman +
-Landlock. matchlock and smolmachines are spiritually close on the
-*policy* side (default-deny net, per-host allowlist) but use microVMs
-instead of containers.
+bot-bottle now defaults to a VM per bottle (Firecracker microVM on KVM
+Linux, Apple Container on macOS) with its own DLP-scanning egress proxy,
+keeping Docker only as a legacy fallback; agent-safehouse uses
+`sandbox-exec`; litterbox uses Podman + Landlock. matchlock and
+smolmachines are close on *both* the policy side (default-deny net,
+per-host allowlist) and — now that bot-bottle has moved off
+containers-by-default — the microVM isolation primitive.
+
+**New closest on agent-tailored policy.** Two governance tools are the
+direct competitors on the "coarse-grained sandbox" axis. **tilde.run**
+has had per-agent DSL RBAC since its launch (though it's hosted SaaS).
+**Microsoft AGT** is the most serious new entrant: per-agent DID
+identity, YAML policy that can allow/deny/sandbox/approve individual tool
+calls per agent, and a dynamic behavioural trust score. It operates at
+the framework tool-call layer, not the network layer — so it's
+complementary to bot-bottle's network/filesystem isolation rather than a
+direct substitute, but on the "does this sandbox know what this agent is
+for?" question it is the most complete answer in the field. OAP's
+pre-action hook pattern achieves similar goals with cryptographic audit
+and a 0% adversarial-attack success rate under a restrictive policy.
+
+**New closest on DX.** **Docker sbx** is the first tool in this set that
+matches bot-bottle on the "one command, dangerously-skip-permissions safe
+by default" DX bar, at microVM isolation strength, with host-side
+credential injection. It is proprietary, preset-based (not role-
+declarative), and cloud-agent-specific, but it directly competes on the
+UX proposition. agent-safehouse was the previous DX peer; Docker sbx
+materially raises the bar.
+
+**New closest on repo-scoped policy.** **Cleanroom** (Buildkite) is the
+first tool to combine microVM isolation with a declarative egress policy
+file — though the policy lives in the repo being sandboxed
+(`cleanroom.yaml`), not in an agent-role manifest. That makes it per-
+repo rather than per-role: the same Cleanroom config applies to any
+agent running in that repo. The distinction matters for bot-bottle's
+use case (one developer running multiple agent *roles* with different
+egress footprints), but for CI/CD use cases Cleanroom is a direct
+alternative.
 
 **Solving a different problem.** tilde.run is hosted SaaS for team /
 production agent pipelines with data-versioned rollback — explicitly
-opposite to bot-bottle's "infrastructure I control" goal. boxlite and
-microsandbox are infrastructure libraries aimed at platform builders
-embedding sandboxes into agent frameworks; they would be a *backend*
-bot-bottle could call, not a competitor to its manifest layer.
-endo-familiar is in a different paradigm entirely: capability passing
-rather than kernel boundaries.
+opposite to bot-bottle's "infrastructure I control" goal. boxlite,
+microsandbox, and CubeSandbox are infrastructure libraries/services aimed
+at platform builders embedding sandboxes into agent frameworks; they
+would be a *backend* bot-bottle could call, not a competitor to its
+manifest layer. endo-familiar is in a different paradigm entirely:
+capability passing rather than kernel boundaries.
 
 ## Borrowable ideas
 
 What bot-bottle already has that the survey suggested as
 differentiators:
-- Default-deny egress with a per-agent allowlist (pipelock).
+- Default-deny egress with a per-agent allowlist (own egress scanner).
 - DLP scanning of outbound traffic.
 - Bottle / agent split (manifest layer above the isolation primitive).
 - gVisor auto-detection on Linux.
 
-Ideas worth considering, without abandoning the Python-stdlib-first / local-Docker
-stance:
+Ideas worth considering, without abandoning the Python-stdlib-first /
+local, single-operator stance:
 
 1. **Per-use SSH key confirmation** (from litterbox). Even with
-   KnownHostKey pinning and pipelock egress, a wrapper SSH agent that
+   KnownHostKey pinning and the egress DLP scanner, a wrapper SSH agent that
    prompts on each key use (e.g. via `osascript` / `notify-send`) would
    catch an agent doing something off-policy with a key it legitimately
    holds. Pure-stdlib, no new deps.
-2. **In-flight secret injection** (from matchlock). Pipelock already
-   does egress allowlisting and DLP; teaching it to *inject* tokens at
+2. **In-flight secret injection** (from matchlock). The egress scanner
+   already does allowlisting and DLP; teaching it to *inject* tokens at
    proxy time so e.g. `GITEA_TOKEN` never appears in the container's
    env would close the "agent reads its own env and exfiltrates" path.
-   Fits the existing pipelock architecture.
-3. **MicroVM backend as an opt-in bottle type** — already on the radar
-   in `stronger-isolation-alternatives.md`. microsandbox, smolmachines,
-   and matchlock all show that libkrun + Apple's
-   Virtualization.framework is ergonomic enough that a
-   `"runtime": "microvm"` field on a bottle is plausible without a heavy
-   stack.
+   Fits the existing egress-proxy architecture.
+3. **MicroVM backend** — ~~on the radar~~ **shipped since this survey.**
+   microVMs are now bot-bottle's default (Firecracker on KVM Linux, Apple
+   Container on macOS); Docker is the legacy fallback. The libkrun / Apple
+   Virtualization.framework ergonomics that microsandbox, smolmachines,
+   and matchlock demonstrated turned out to be enough to make it the
+   default rather than an opt-in.
 
 Not worth borrowing: the SDK-first programmatic API style of boxlite /
 microsandbox (cuts against the declarative-manifest stance), and the
@@ -230,3 +469,176 @@ hosted-SaaS dashboard model of tilde.run (cuts against the
 - The `superradcompany/microsandbox` URL in the original prompt
   redirects to `microsandbox/microsandbox`; the surveyed project is the
   same.
+- CubeSandbox performance/scale numbers (<60ms cold start, <5MB/instance,
+  2,000+ sandboxes per 96-vCPU host) are the project's own launch claims,
+  not independently verified here.
+
+## Addendum 2026-07-18 — CubeSandbox and the positioning read
+
+CubeSandbox (Tencent Cloud, Apache 2.0, ~10.4k stars, HN launch
+[#47863430](https://news.ycombinator.com/item?id=47863430)) is the first
+project in this survey to combine, in one open-source stack, everything
+bot-bottle treated as its differentiator:
+
+- **Egress custody (connection level)** — default-deny domain allowlist
+  (L7 domain/SNI filtering), instant block on unauthorized egress,
+  per-sandbox traffic tokens, full audit logs of destinations (eBPF
+  virtual switch, "CubeVS"). This matches bot-bottle's egress scanner at
+  the *connection level*, productized — see the one thing it does **not**
+  match, below.
+- **Credential custody** — a vault where keys "never enter the sandbox,
+  model context, or logs." This is the in-flight-injection idea from
+  matchlock, but as a first-class feature, and it's exactly the
+  cross-vendor "egress audit + custody" wedge the monetization
+  positioning treats as the one defensible moat.
+- **Isolation on par with bot-bottle's current default** — a dedicated
+  guest kernel per sandbox (RustVMM/KVM). bot-bottle now defaults to the
+  same class of boundary (Firecracker microVM / Apple Container), so this
+  is parity, not an edge; CubeSandbox's remaining edge is running that
+  per-kernel isolation multi-tenant at scale on one host.
+
+The one axis CubeSandbox does **not** cover — and where bot-bottle stays
+distinctive:
+
+- **Content DLP on *authorized* channels.** CubeSandbox's egress control
+  is connection-level: it decides *whether* a destination is allowed and
+  logs it, and its vault keeps *injected* credentials out of the sandbox
+  entirely. Neither inspects the *payload* of traffic to an allowed
+  destination. So an agent that exfiltrates over a permitted channel —
+  pasting a repo's contents, an agent-derived secret, or PHI into an
+  allowed API/domain — is not caught by CubeSandbox. bot-bottle's own
+  egress DLP scanner does scan that: response + websocket content against
+  the resolved per-flow config, with per-bottle token redaction (see
+  recent egress commits). The vault
+  approach is arguably *stronger* for the specific case of pre-known
+  injected credentials (they can't leak if they were never present), but
+  it is not a substitute for content inspection of everything else.
+
+**Long-running posture — a sharper axis than raw isolation.** E2B and
+CubeSandbox are *ephemeral-per-task* by design; a long-running agent is an
+architected pattern on top, not the default. E2B: 5-minute default
+timeout, continuous runtime tier-capped (~1h Hobby / ~24h Pro), duration
+achieved via **pause/resume** (preserves filesystem + memory + processes;
+reconnect by sandbox ID via `Sandbox.connect()`; resume resets the timeout
+to 5 min; auto-pause via `on_timeout: "pause"`). CubeSandbox mirrors this
+(E2B drop-in) with first-class auto pause/resume and hundred-ms
+checkpoint/fork — and, self-hosted, sets its own timeout policy with no
+vendor tier caps. bot-bottle inverts the model: a bottle is **persistent,
+named, and supervised by default** — long-running *is* the default, not a
+session-management loop over pause/resume. smolmachines is the other
+persistent-by-default project in this set. For anyone building agents that
+run for hours/days, this posture difference matters more than the
+isolation primitive.
+
+**DX — the "run Claude yolo-style" bar.** The reason `claude
+--dangerously-skip-permissions` is so widely used is DX: it's one command
+and the agent just goes. The bottle thesis is to make a *sandboxed* run
+that easy — `start <agent>` builds the image on first run and drops you
+into an interactive Claude session that already has
+`--dangerously-skip-permissions` on by default
+(`contrib/claude/agent_provider.py`), with the sandbox as the guardrail
+instead of per-action prompts. On this axis the field splits cleanly:
+- **Wrappers around the agent** (as-easy-as-native): bot-bottle and
+  **agent-safehouse** (`safehouse claude --dangerously-skip-permissions`).
+  These *are* the run-Claude experience. agent-safehouse is the real DX
+  peer — but it's macOS-only Seatbelt, single-run, and doesn't address
+  network egress; bot-bottle adds VM-grade isolation, egress DLP, and
+  persistent/parallel bottles across macOS + Linux.
+- **Libraries / services** (you build the run yourself): boxlite,
+  microsandbox, CubeSandbox, E2B. These hand you an SDK or a cluster and
+  expect you to wire the agent in — powerful for platform builders,
+  heavyweight for "just run Claude on my laptop." microsandbox's MCP/Skills
+  angle is *sandbox-as-a-tool the agent calls*, which is the inverse of
+  wrapping the agent.
+- **In between:** litterbox (wizard + build, Linux only), smolmachines
+  (SSH into a named machine), matchlock (run a command in a VM).
+
+So DX is a genuine bot-bottle differentiator, and the only project that
+matches it (agent-safehouse) does so with materially weaker isolation and
+no egress story. "As easy as native yolo, but actually sandboxed" is a
+defensible one-liner.
+
+Why it still doesn't collide head-on:
+
+1. **Shape.** CubeSandbox is a *multi-tenant service for platform
+   builders* (drop-in E2B replacement, SDK-driven, 2,000 sandboxes on a
+   box). bot-bottle is a *single-operator, declarative-manifest tool for
+   the infrastructure I run*. Different buyer, different ergonomics — no
+   JSON manifest, no bottle/agent split, no "one command on my laptop."
+2. **Backend, not competitor.** Like boxlite/microsandbox, CubeSandbox is
+   something bot-bottle could sit *on top of* — a `"runtime": "microvm"`
+   or `"runtime": "cubesandbox"` backend under the manifest layer — while
+   keeping the manifest, the bottle/agent split, and the local,
+   single-operator default.
+
+Why it matters anyway:
+
+- The "nobody else bundles connection-level egress allowlist + audit +
+  in-flight credential custody" line is **no longer true for the
+  primitive** — a well-funded, 10k-star open-source project now ships it.
+  But **content DLP on authorized channels is still not matched** (see
+  above), and neither is the *layer above* the primitive (declarative
+  manifest, cross-vendor orchestration, operator UX, the
+  phone-control/dashboard north star). Those two — outbound-payload DLP
+  and the orchestration layer — are where the defensible ground now sits;
+  the connection-level allowlist + vault mechanism, on its own, is no
+  longer differentiating. Revisit the monetization open/paid line with
+  that in mind.
+- Worth a closer look at **how** CubeSandbox does credential injection
+  and per-sandbox egress tokens (eBPF virtual switch vs. bot-bottle's
+  mitmproxy egress proxy) before the next iteration of bot-bottle's
+  in-flight-secret feature — see borrowable idea #2 above.
+
+## Addendum 2026-07-18 (second pass) — agent-tailored policy landscape
+
+The second-pass question was: how novel is bot-bottle's per-agent,
+role-tailored sandbox relative to the expanded field?
+
+**The short answer:** on the isolation + network + role-tailoring
+combination, bot-bottle remains the only tool in this set. On
+role-tailored *policy at the tool-call level*, Microsoft AGT and OAP are
+the most complete answers, but they don't provide isolation; they
+complement rather than substitute.
+
+**The competitive picture by axis:**
+
+- *Agent-tailored egress (declarative, per-role)* — bot-bottle and
+  tilde.run. Cleanroom is per-repo, not per-role. Everyone else is
+  per-session or not addressed.
+- *Agent-tailored tool-call policy (declarative, per-agent identity)* —
+  Microsoft AGT (YAML policy + DID identity + trust score), OAP
+  (declarative policy rules + cryptographic audit). Neither provides
+  network/filesystem isolation.
+- *Composable policy (role overlays)* — bot-bottle (`extends:`). No
+  other tool surveyed supports composable role-policy inheritance.
+- *Isolation + DX (one-command safe yolo)* — bot-bottle and Docker sbx.
+  Docker sbx is proprietary, preset-based, and cloud-agent-specific;
+  it's the first DX-class competitor at microVM isolation strength.
+
+**What the HN "coarse-grained" complaint maps to:** The complaint is
+that a VM isolates the filesystem but doesn't know if the agent
+*should* be sending an email. bot-bottle's bottle/agent split is a
+structural answer to this: the bottle manifest declares exactly what
+the role can reach, and the sandbox enforces it at the network layer.
+Microsoft AGT is the most complete answer at the semantic/tool-call
+layer. The gap both leave open is *intent classification* — knowing
+whether a permitted action is consistent with the agent's actual task.
+See `hn-agent-safety-discourse-july-2026.md` for the blast-radius
+analysis.
+
+**Borrowable from new tools:**
+
+- **Microsoft AGT's trust-score decay** — privilege that reflects
+  observed behaviour rather than static provisioning. Applied to
+  bot-bottle: a bottle that has triggered DLP alerts or supervise holds
+  could auto-downgrade its network preset, or flag the session for
+  closer review. Fits the existing supervise-server architecture.
+- **Docker sbx's live network TUI** — real-time per-session view of
+  allowed and blocked outbound connections with point-and-click
+  allow/block. `cli.py supervise` is the right surface; adding a
+  live-connections panel would directly address the "I can't see what
+  the agent is doing" gap without any backend changes.
+- **OAP's cryptographic audit chain** — Ed25519-signed, hash-chained
+  audit records. Currently bot-bottle logs egress decisions but doesn't
+  chain them. A tamper-evident audit record per session would be useful
+  for the compliance use case the CubeSandbox positioning targets.

docs/research/hn-agent-safety-discourse-july-2026.md

@@ -0,0 +1,345 @@
+# HN discourse on agent sandbox safety — June/July 2026
+
+A survey of community opinion and notable security disclosures on Hacker
+News and adjacent sources over June–July 2026. The question: what does
+the current discourse say about whether sandboxes are sufficient for
+agentic AI safety, and where does bot-bottle land against the issues
+being raised?
+
+Research conducted 2026-07-18.
+
+## Summary
+
+The past month marks a turning point in community opinion. Earlier in
+2026, the debate was mostly "which sandbox tool is best?" By June–July,
+a cascade of critical CVEs and novel attack classes has shifted the
+framing to "sandboxes are not enough — what else do you need?" The
+attacks that drove this shift are structurally distinct: most route
+through legitimate, trusted channels (Sentry issues, MCP descriptions,
+README files) rather than exploiting the isolation boundary directly.
+
+bot-bottle's architecture holds up well against the direct-escape class
+(Firecracker/Apple Container default backends, credentials never in the
+agent's env, harness entirely on the host). The remaining gap is prompt
+injection — attacker-controlled data interpreted as model instructions.
+Egress controls and prompt injection defenses are orthogonal: egress
+limits what the agent can *send out*; injection is about what it is
+*told to do*. The two don't substitute for each other. Inside a tightly-
+egressed sandbox a successful injection can't exfiltrate to unknown
+hosts, but it can still corrupt the work product, push malicious commits
+past a secret scanner, or use allowlisted channels for exfiltration.
+Those residual risks are addressed below.
+
+## The sandboxing boom sets the stage
+
+The preceding months generated a wave of sandbox tooling. A March 28
+Ask HN thread
+([#47444917](https://news.ycombinator.com/item?id=47444917)) catalogued
+the explosion: E2B, AIO Sandbox, AgentSphere, Yolobox, Exe.dev,
+AgentFence, DenoSandbox, Capsule (WASM), ERA, Vibekit, Daytona, Modal,
+Nono, and more — all launched within roughly 12 months. A parallel March
+9 thread ([#47185250](https://news.ycombinator.com/item?id=47185250))
+surveyed what developers were actually deploying: "containers or YOLO"
+dominated. The honest community mood was that most teams hadn't solved
+this and were shipping anyway.
+
+## The June–July attack cascade
+
+Six attack patterns broke in quick succession. Together they form the
+argument that the community's framing was wrong: the threat model for
+agents isn't just "code that escapes its container" — it's also prompt
+injection, where attacker-controlled data is interpreted as model
+instructions regardless of whether any isolation boundary was crossed.
+Sections 2–4 below are all the same attack class; the "trusted channel"
+label describes the delivery vector, not a different threat.
+
+### 1. Sandbox escape CVEs (DuneSlide, CVE-2026-39861)
+
+Cato AI Labs disclosed **DuneSlide** (CVE-2026-50548/50549, CVSS 9.8),
+a pair of flaws in Cursor 2.x. CVE-2026-50548 abuses the sandbox's
+`working_directory` parameter to point writes at system files; CVE-26-50549
+exploits a symlink-resolution fallback that fails open. Both start with
+a prompt injection and end in sandbox escape — and Cato's framing was
+blunt: "each CVE defeats a different guardrail; the problem is
+structural, not a string of one-offs."
+
+Claude Code's own sandbox had a similar escape this year:
+**CVE-2026-39861** (symlink flaw). The CurXecute/MCPoison/CVE-2026-26268
+chain from Cursor added a poisoned Slack message, a swap-after-approval
+MCP config, and a Git hook as three more entry points in the same
+attack class.
+
+All patched, but the pattern holds: any application-level sandbox that
+takes attacker-influenced values as path parameters is reachable from a
+prompt injection.
+
+### 2. Prompt injection via MCP data (Agentjacking)
+
+Tenet's "Agentjacking" technique planted a fake bug report in Sentry's
+MCP output. When an agent queries Sentry to fix open issues, the
+malicious event is rendered as structured content visually
+indistinguishable from a real Sentry event, and the agent executes the
+embedded instructions with the developer's full privileges. Hit rate
+across Claude Code and Cursor: **85%**. The route is entirely through a
+legitimately-authorized MCP channel — no isolation boundary is crossed;
+the injection arrives inbound through a channel the sandbox explicitly
+trusts.
+
+The Cloud Security Alliance's summary: treat observability, bug-report,
+and integration data as **untrusted agent input**, not neutral
+development metadata.
+
+### 3. README-embedded prompt injection
+
+A July disclosure showed malicious instructions hidden in `README.md`
+— a file that receives no trust prompt and requires no elevated access.
+When asked point-blank whether the repo held hidden instructions, both
+Claude Sonnet 4.6 and GPT-5.5 said no. A payload written for Sonnet
+4.6 transferred unchanged to Sonnet 5, Opus 4.8, and GPT-5.5. The
+attack surface is every repo an agent is asked to work in.
+
+### 4. Prompt injection via MCP tool descriptions
+
+Microsoft research (June 30) showed that attacker-controlled MCP tool
+description fields can silently redirect agent behavior. The injection
+is embedded in metadata the model reads during tool selection — before
+any sandbox enforcement or egress check runs, and entirely on the
+inbound path that egress controls cannot touch.
+
+### 5. MCP STDIO command injection (10 CVEs)
+
+OX Security disclosed a systemic command injection class in Anthropic's
+MCP protocol, covering 10 CVEs across multiple coding agents. The
+Windsurf case (CVE-2026-30615): processing attacker-controlled HTML
+causes the agent to auto-register a malicious MCP STDIO server and
+execute arbitrary commands with no further user interaction.
+
+### 6. LiteLLM gateway compromise (CVE-2026-40217, CVE-2026-42271)
+
+CVE-2026-40217 exposes LiteLLM's guardrail sandbox via `exec()` with no
+source filtering. CVE-2026-42271 (exploited in the wild, added to CISA's
+KEV catalog) lets callers spawn subprocesses through MCP preview
+endpoints. The threat extends to any agent routed through a compromised
+LiteLLM proxy: the proxy can swap model responses for forged tool calls
+in transit, giving the attacker a reverse shell from the developer's
+machine.
+
+## HN community opinion clusters
+
+**"Move enforcement to the kernel, not the app"** — the Nono Show HN
+([#46849615](https://news.ycombinator.com/item?id=46849615)) and a
+kernel-sandbox thread
+([#47066574](https://news.ycombinator.com/item?id=47066574)) both argued
+that application-layer sandboxes are inherently bypassable by the code
+they're sandboxing. The academic framing, from *Red-Teaming the Agentic
+Red-Team* ([arXiv 2606.24496](https://arxiv.org/pdf/2606.24496)):
+"enforcement should occur at the OS level via the kernel refusing system
+calls that violate policy at runtime — not pre-execution argument
+validation in tool calls."
+
+**"The harness belongs outside the sandbox"** — a May thread
+([#47990675](https://news.ycombinator.com/item?id=47990675)) converged
+on clean architectural separation: harness in one VM, tool execution in
+another. Top comment: "having the harness in one VM, and tool use applied
+to user data in another, is about as safe as you can be at present."
+Several replies described a hypervisor-like policy layer — sitting outside
+both VMs — as the right long-term model.
+
+**"Sandboxes are too coarse-grained"** — a Feb thread
+([#47006445](https://news.ycombinator.com/item?id=47006445)) argued
+that VMs don't answer the real question: knowing whether an agent
+*should* be sending an email or making a transaction. "Everything's just
+in the same big box." This framing picked up traction through June–July
+as the trusted-channel attacks dominated.
+
+**"MCP's trust model is the real problem"** — the month's recurring
+theme. MCP by design gives agents access to authorized external services.
+Once a trusted channel delivers a malicious payload, filesystem sandboxing
+is irrelevant. The community call: treat all MCP tool metadata and return
+values as untrusted input subject to policy validation before ingestion,
+and disable automatic MCP server loading from untrusted repositories.
+
+## How bot-bottle addresses these issues
+
+### What it covers well
+
+**Direct sandbox escape (CVEs, container breakout)**
+
+bot-bottle's default backends are Firecracker microVM (KVM Linux) and
+Apple Container (macOS). Both run the agent in a separate VM with a
+dedicated kernel — the container-escape CVE class (Dirty Pipe, runc
+escapes, DuneSlide's path-parameter abuse) requires escaping a real
+hypervisor boundary, not just a namespace. On the legacy Docker backend,
+gVisor auto-detection provides a userspace syscall barrier for hosts where
+neither KVM nor Apple Container is available.
+
+The bot-bottle process itself runs entirely on the host, outside the VM.
+This is the "harness outside the sandbox" architecture the HN thread
+converged on as best practice. The bottle manifest, egress rules, and
+secrets never enter the agent VM.
+
+**Credential theft on sandbox escape**
+
+Even on a successful VM/container escape, the agent has nothing useful
+to steal. Credentials are injected in-flight by the gateway proxy
+(`auth.scheme` / `auth.token_ref` in the egress route config) — `printenv`
+inside the agent shows proxy URLs only. The git-gate similarly holds the
+upstream SSH credential on the host; the agent pushes through a
+gitleaks-scanned daemon that forwards clean refs upstream. An escaped
+agent gets the host filesystem, not the keys.
+
+**Orphaned-agent credential risk**
+
+bot-bottle is explicitly ephemeral: when the agent exits, `cli.py` tears
+down every gateway and both networks — nothing persists between runs. The
+agent never holds credentials, so there is nothing to orphan.
+
+**MCP config redirection / STDIO auto-registration**
+
+The trust boundary at `$HOME` means bottles live only under
+`~/.bot-bottle/bottles/` — a cloned repo cannot add egress routes or
+redirect env vars to attacker hosts (the design rationale is in
+`docs/prds/0011-per-file-md-manifest.md`). Auto-registering a malicious
+MCP STDIO server from within the agent is still sandboxed by the VM, and
+any outbound calls from that server must pass the egress allowlist and
+outbound DLP scanner.
+
+**Per-agent role tailoring (the "coarse-grained sandbox" complaint)**
+
+The Feb 2026 HN thread that argued "sandboxes are too coarse-grained"
+was pointing at a real gap: a VM isolates the filesystem but doesn't
+know whether an agent *should* be sending email or calling an external
+API. bot-bottle's bottle/agent split is a structural answer at the
+network layer — the bottle manifest declares exactly what each role can
+reach (which hosts, which paths, which HTTP methods), and the egress
+scanner enforces it. A `gitea-dev` bottle that only lists
+`gitea.dideric.is` and `api.anthropic.com` structurally cannot send
+email or reach AWS, not because the model was told not to, but because
+those routes don't exist.
+
+The `extends:` composition model means provider-level policy (the Claude
+auth route) lives in one base bottle and role-specific overlays are
+stacked on top — no duplication, and changing the base propagates to all
+derived roles.
+
+Competitive position on this axis (from `agent-sandbox-landscape.md`):
+
+| Tool | Agent-tailored policy |
+|---|---|
+| **bot-bottle** | Yes — declarative per-role manifest; `extends:` composition; egress + credentials scoped to role |
+| **tilde.run** | Yes — per-agent DSL RBAC (allow/deny/approve per action/repo/agent), but hosted SaaS |
+| **Microsoft AGT** | Yes — YAML policy + per-agent DID + trust score, but tool-call level only (no network isolation) |
+| **OAP** | Yes — declarative pre-action policy + cryptographic audit, but no isolation |
+| **Cleanroom** | Partial — per-repo `cleanroom.yaml`, not per-role |
+| **Docker sbx** | No — network presets only |
+| **Anthropic srt** | No — programmatic per-invocation |
+| **matchlock / smolmachines / microsandbox** | No |
+| **agent-safehouse** | Partial — per-agent Seatbelt profiles; no egress |
+
+Two takeaways: bot-bottle and tilde.run are the only isolation tools
+with declarative role-tailored policy; Microsoft AGT and OAP are the
+closest competitors on role-tailoring but operate at the tool-call layer
+without network/filesystem isolation — complementary, not substitutes.
+
+**Outbound exfiltration (any injection class)**
+
+Whatever triggers the agent — README injection, Agentjacking, MCP
+description poisoning — the final step in most attacks is exfiltration.
+bot-bottle's egress allowlist is default-deny with a per-bottle host
+allowlist; unknown hosts get a hard 403. Outbound DLP scanning
+(`outbound_detectors: [token_patterns, known_secrets]`) catches tokens
+and secrets in outbound bodies; the `supervise` policy (default for
+manifest routes) holds the request for operator approval rather than
+silently blocking it. Together these limit what a successful injection
+can *do* even if it succeeds at the model layer.
+
+**LiteLLM / compromised-proxy attacks**
+
+bot-bottle does not use LiteLLM. The model API route (e.g.
+`api.anthropic.com`) is an auto-injected provider route on the egress
+allowlist; the agent dials the gateway, not the model API directly.
+A compromised third-party proxy is not in the architecture.
+
+### Where it is weaker
+
+**Prompt injection**
+
+Egress controls and prompt injection defenses are orthogonal. Egress
+limits what the agent can *send out* (outbound leg); prompt injection
+is about what attacker-controlled data *tells the agent to do* (inbound
+leg). The two don't substitute for each other and must be treated
+separately.
+
+The inbound DLP scanner (`inbound_detectors: [naive_injection_detection]`)
+is the only runtime defense against injection arriving through allowlisted
+channels — Sentry MCP responses, MCP tool descriptions, README content.
+It is explicitly pattern-matching and will not catch a sufficiently
+crafted payload. There is no semantic / intent-level gate between what
+the model decides and what the agent executes.
+
+**Blast radius within the permitted scope**
+
+Inside a tightly-egressed sandbox a successful injection can't
+exfiltrate to unknown hosts, but it still has real options:
+
+- *Work product corruption.* The agent can modify, delete, or backdoor
+  files in the working directory. This is within its permitted scope;
+  egress controls have nothing to say about it.
+
+- *Malicious commits past the git-gate.* The git-gate scans outbound
+  refs for secrets (gitleaks), not for semantic code intent. A prompt-
+  injected agent can commit subtly malicious code — logic bombs,
+  backdoored auth paths, code that exfiltrates data through the
+  application's own HTTP clients at runtime — that looks clean to a
+  secret scanner.
+
+- *Exfiltration through allowlisted channels.* If an attacker knows or
+  can predict what hosts are in the egress allowlist, those channels are
+  available for exfiltration. A GitHub remote being allowlisted means
+  "push to an attacker-controlled fork" is viable. A logging endpoint
+  being allowlisted means structured data can leave through it. The
+  outbound DLP scanner catches credential tokens and known secrets but
+  not arbitrary business data.
+
+- *Dependency installation within the sandbox.* An agent that runs
+  `npm install` or `pip install` on attacker-specified packages executes
+  code inside the sandbox with the same capabilities the agent has:
+  filesystem access, tool calls, calls to allowlisted hosts. Supply chain
+  injection via package names is in the same injection family, triggered
+  by the same prompt-injection path.
+
+### What would close the remaining gaps
+
+The blast-radius risks above point at two distinct mitigations that
+don't yet exist in bot-bottle:
+
+- *Outbound intent classification.* The egress addon today scans
+  outbound request content for token patterns. What it lacks is
+  awareness of context — it can't distinguish "agent is pushing a
+  legitimate commit" from "agent was injected and is pushing a backdoor."
+  The `supervise` policy is already the right shape for human-in-the-loop
+  review on sensitive outbound actions; extending it with context from
+  the agent's recent tool calls (what files were touched, what was the
+  triggering task) would narrow the gap.
+
+- *Semantic code review on git push.* gitleaks is the wrong tool for
+  catching injected logic. A review step on outbound commits — even a
+  simple diff summary surfaced in `cli.py supervise` before the push is
+  forwarded — would close the malicious-commit path without requiring
+  the agent to be fully trusted.
+
+## Sources
+
+- [Ask HN: The new wave of AI agent sandboxes? (Mar 2026)](https://news.ycombinator.com/item?id=47444917)
+- [OK, let's survey how everybody is sandboxing AI coding agents (Mar 2026)](https://news.ycombinator.com/item?id=47185250)
+- [The agent harness belongs outside the sandbox (May 2026)](https://news.ycombinator.com/item?id=47990675)
+- [Show HN: Nono – Kernel-enforced sandboxing for AI agents (Feb 2026)](https://news.ycombinator.com/item?id=46849615)
+- [Kernel-enforced sandbox for AI agents, MCP and LLM workloads (Feb 2026)](https://news.ycombinator.com/item?id=47066574)
+- [Sandboxes will be left in 2026 (Feb 2026)](https://news.ycombinator.com/item?id=47006445)
+- [Critical Cursor Flaws / DuneSlide – The Hacker News](https://thehackernews.com/2026/07/critical-cursor-flaws-could-let-prompt.html)
+- [Agentjacking Attack – The Hacker News](https://thehackernews.com/2026/06/agentjacking-attack-tricks-ai-coding.html)
+- [Friendly Fire: AI Agents Built to Catch Malicious Code – The Hacker News](https://thehackernews.com/2026/07/friendly-fire-ai-agents-built-to-catch.html)
+- [Microsoft Warns Poisoned MCP Tool Descriptions – The Hacker News](https://thehackernews.com/2026/06/microsoft-warns-poisoned-mcp-tool.html)
+- [MCP STDIO Command Injection Advisory – OX Security](https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/)
+- [LiteLLM Vulnerability Chain – The Hacker News](https://thehackernews.com/2026/06/litellm-vulnerability-chain-lets-low.html)
+- [Red-Teaming the Agentic Red-Team (arXiv 2606.24496)](https://arxiv.org/pdf/2606.24496)

docs/research/malicious-commit-scanning.md

@@ -0,0 +1,204 @@
+# Scanning for malicious commits at the git-gate — and whether the semantic layer is a paid feature
+
+A sandboxed agent's most durable way out isn't live exfiltration — it's the
+**artifact it pushes**. Code committed through the git-gate leaves the
+sandbox and runs *later*, in CI or production, with none of the bottle's
+isolation or egress control. gitleaks covers *secrets* on that channel
+(see [`git-gate` pre-receive](../../bot_bottle/git_gate_render.py)); this
+note asks the harder question — can we scan for *malicious code* in
+commits — and the product question the answer raises: is the semantic
+(LLM) layer a decent paid feature on "git egress," alongside a paid
+human-review console for the supervise flow?
+
+## Summary
+
+- **No scanner reliably detects malicious code.** It's undecidable in
+  general and adversarial in practice — a capable agent hides a backdoor
+  from any pattern scanner (and from human reviewers). So the frame is not
+  "detect malice"; it's **raise the cost, cover the common/obvious cases,
+  and put a human on the high-leverage ones.**
+- **Layered, ranked by ROI:** (1) supply-chain / malicious-dependency
+  scanning (Socket, OSV-Scanner, GuardDog) — highest ROI, best git fit;
+  (2) fast heuristic / obfuscation / dangerous-sink scanning
+  (Semgrep-on-diff) — gitleaks-shaped; (3) **risk-based human gating** via
+  the existing supervise plane on high-risk file classes; (4) **LLM
+  semantic diff-review** as a best-effort backstop for novel backdoors.
+- **Architecture:** the git-gate's `pre-receive` is *synchronous* (it
+  blocks the push), so run only fast scanners inline and push heavy
+  analysis (CodeQL, full LLM review) to an **async post-push CI lane** that
+  can't hold the developer hostage.
+- **Monetization verdict: yes — but the paid unit is the *governed
+  git-egress review bundle*, not the raw scanner.** The pre-receive
+  mechanics and the fast OSS scanners stay free (open-core social
+  contract, same as gitleaks). The sell is the control-plane workflow
+  around them: **managed semantic review + a web-console human-review flow
+  + RBAC + audit retention + cross-run policy.** A bare per-push scan is
+  borderline "single-run = free" by our own heuristic, and its detection
+  accuracy is not something to sell on. Sell the *workflow and the
+  governance*, not the detector.
+
+## The threat: temporal escape
+
+This is the one containment fundamentally can't close. bot-bottle's VM
+isolation + egress DLP shrink *live* exfiltration, but the git push
+channel launders a payload past all of it: the agent commits code that
+looks benign at review time and does its work later, outside the
+perimeter — the "blind-XSS"-style temporal escape that both the "3 reasons
+sandboxing won't secure your AI agent" talk and Embroidery's "In sandboxes
+we (shouldn't) trust" call the sharpest gap. gitleaks gates *secrets* on
+this channel; malicious *code* is the uncovered surface, and it's the
+behavioral half of the isolation-vs-behavioral split noted in
+[`agent-sandbox-landscape.md`](agent-sandbox-landscape.md).
+
+## Why it's fundamentally hard
+
+Distinguishing malicious from legitimate code is undecidable in general
+and adversarial in the specific. The underhanded-code contests show humans
+hide backdoors from *human reviewers*; a scanner reduces to regex/dataflow
+heuristics a motivated agent writes around. So "catch malicious commits"
+is the wrong success metric. The right ones: *how much did we raise the
+cost*, *did we cover the common and the obvious*, and *did a human see the
+dangerous changes before they left*.
+
+## Landscape of git-pluggable approaches (ranked by ROI)
+
+1. **Supply-chain / malicious-dependency scanning — highest ROI.** The
+   most likely agent-introduced threat is a malicious or typosquatted
+   *dependency*, not a hand-written backdoor — and it's far more
+   detectable. **Socket** (behavioral analysis of packages: install
+   scripts, obfuscation, unexpected network/fs access, typosquats) is the
+   closest thing to malicious-intent detection that actually works;
+   **OSV-Scanner** (known-bad DB, fast) and **GuardDog** (PyPI/npm
+   heuristics) complement it. Trigger only when a commit changes a
+   lockfile/manifest → block or escalate.
+2. **Fast heuristic / obfuscation / dangerous-sink scanning.** **Semgrep**
+   with a curated ruleset, scanning the *diff* (fast enough for
+   pre-receive). Catches obfuscation (`eval(atob(...))`, base64/hex blobs,
+   minified source), exfil primitives (`curl | sh`, reverse shells,
+   DNS-exfil), and dangerous sinks (`shell=True`, `pickle.loads`, unsafe
+   `yaml.load`, `child_process.exec`). Tune noisy rules to *escalate*, not
+   block.
+3. **Risk-based human gating — the most robust control, already
+   half-built.** The git-gate already escalates `# gitleaks:allow`
+   findings to a supervisor proposal and blocks the push pending approval
+   (`supervise_gitleaks_allow`, [`git_gate_render.py`](../../bot_bottle/git_gate_render.py)).
+   Extend the same flow to **high-risk file classes**: any commit touching
+   CI/build/deploy scripts, auth/crypto code, egress config, or
+   adding/changing dependencies → route to `./cli.py supervise`. This is
+   attribution/policy, not detection, and it's the strongest thing here —
+   a human on exactly the temporal-escape surfaces.
+4. **LLM semantic diff-review — the behavioral backstop.** The only
+   approach with a shot at a novel, benign-looking backdoor: a model
+   reviews the diff for a backdoor / exfiltration / obfuscated payload /
+   dangerous behavior. This is the behavioral-detection thesis applied to
+   the *artifact* (the diff) rather than the agent's reasoning. Real
+   caveats: false positives and negatives, latency/cost, it can be
+   prompt-injected by hostile content *in the diff itself*, and it invites
+   over-trust. Useful, never a guarantee.
+
+**Not for this job:** CodeQL, Trivy, Grype, Bandit. They find *known
+vulns and insecure patterns* (bugs), not deliberate backdoors, and the
+powerful ones (CodeQL taint) need a build + database — too heavy for a
+synchronous gate. They belong in the async CI lane if at all.
+
+## Fit into bot-bottle's git-gate
+
+The `pre-receive` hook today is: gitleaks-scan each ref → escalate
+`# gitleaks:allow` findings to supervise → forward to upstream
+([`git_gate_render.py`](../../bot_bottle/git_gate_render.py)). The
+additions slot in cleanly:
+
+- **Inline (fast), before forward:** a dep-scan phase (on manifest/lockfile
+  change) and a Semgrep-diff phase. Findings block or open a supervise
+  proposal, same shape as gitleaks.
+- **New supervise tool types** alongside the existing
+  `egress-block/allow`, `gitleaks-allow`, `egress-token-allow`
+  ([`supervise_types.py`](../../bot_bottle/supervise_types.py)) — e.g. a
+  `commit-review` proposal for risky-file-class gating and for semantic
+  review. The supervise plane is already the right abstraction; this is
+  another *producer* feeding it, and [`supervise_server.py`](../../bot_bottle/supervise_server.py)
+  (JSON-RPC) is already the console backend.
+- **Async lane (heavy):** full LLM review + any CodeQL run out of band
+  after the push, feeding the same review/audit surface, so the
+  synchronous gate stays fast.
+
+## The product question: paid feature on git egress?
+
+Restating the open-core line bot-bottle runs on: *give away the
+sandbox/runtime, charge for the control plane; single-run/single-node =
+free, cross-run aggregation + central enforcement + identity/fleet = paid;
+the moat is uniform egress audit + secret custody + policy across
+untrusted agents.*
+
+Against that line, the split is clean:
+
+**Free (OSS runtime — the trust funnel):**
+- the `pre-receive` gate mechanics and gitleaks;
+- wiring the OSS scanners (Socket CLI / OSV-Scanner / Semgrep);
+- the CLI supervise flow.
+Keeping the raw scanners free is the same social contract as gitleaks and
+preserves the bottom-up distribution funnel.
+
+**Paid (the governed git-egress bundle — the control plane):**
+- **Managed semantic diff-review** — hosted inference + a curated,
+  maintained malicious-pattern/policy set. This is *capability* (metered),
+  not *insurance* — the thing individuals actually pay for. Position it as
+  **governed code-egress review**, not "we resell inference" (the
+  monetization notes explicitly warn against reselling compute).
+- **The web-console supervise/review flow — the strongest anchor.** Turn
+  the CLI `./cli.py supervise` approval into a real review surface:
+  rendered diff + finding context, approve/reject, **who-approved audit
+  trail, RBAC on approvers, mobile/phone-control** (ties to the
+  dashboard/vault north star). This is "central enforcement +
+  identity/fleet = paid" almost verbatim — and it generalizes across
+  *every* supervise proposal (egress block/allow, gitleaks-allow,
+  commit-review), so it's worth building for the whole plane, with the
+  semantic check as one producer.
+- **Cross-run governance:** fleet-wide policy for what escalates,
+  review-decision history/search/export, and drift alerts.
+
+**Why it fits the moat rather than bolting on:** a git push *is* an egress
+channel. A semantic review + human approval + audit on it extends the
+uniform "egress audit + custody + policy across untrusted agents" wedge to
+**code artifacts** — the same product, applied to the one channel gitleaks
+only half-covers. That's on-moat, not a detour.
+
+**The honest nuance (don't oversell):** a bare per-push LLM scan is
+arguably *free* by the single-run heuristic, and its detection accuracy is
+not defensible to charge for. The paid value is the **governance around
+it** — the console, RBAC, audit retention, cross-run policy — plus the
+managed capability. Sell the *review-and-approve-and-audit workflow*; let
+the detector be explicitly best-effort. And per the monetization
+guardrail, the "anti-corporate" free crowd must not veto these team
+features: the review console + RBAC + audit *are* the monetization.
+
+## Recommendation
+
+1. **Land the free layer first.** Add the dep-scan and Semgrep-diff phases
+   to `pre-receive`, and extend supervise to risky-file-class gating —
+   reuses existing machinery, immediate value, stays OSS.
+2. **Build the supervise web console** over `supervise_server`'s JSON-RPC
+   (already the Phase-1 move in the monetization path). This is the paid
+   anchor and it serves *all* proposal types, not just commit review.
+3. **Add managed semantic diff-review as a paid producer** feeding that
+   console — "governed code-egress review," metered, explicitly
+   best-effort on detection.
+4. **Don't oversell detection.** Market the workflow (review + approve +
+   audit) and the cross-run policy/RBAC, where the value is real and
+   defensible; keep the raw scanners open.
+
+## Sources / references
+
+- [`agent-sandbox-landscape.md`](agent-sandbox-landscape.md) — the
+  egress-DLP gap and isolation-vs-behavioral framing.
+- Git-gate internals: [`git_gate_render.py`](../../bot_bottle/git_gate_render.py),
+  [`supervise_types.py`](../../bot_bottle/supervise_types.py),
+  [`supervise_server.py`](../../bot_bottle/supervise_server.py).
+- External tools: Socket (socket.dev), OSV-Scanner (google/osv-scanner),
+  GuardDog (DataDog/guarddog), Semgrep (semgrep/semgrep).
+- Threat framing: "3 reasons sandboxing won't secure your AI agent"
+  (youtube TsYDazwHJ6U); Embroidery, "In sandboxes we (shouldn't) trust."
+- The authoritative monetization/positioning analysis (the open-core line,
+  the wedge, single-run-free/cross-run-paid) lives in the **separate
+  `bot-bottle-console` repo**, not this one — cited here from memory, not
+  linked.

scripts/tracker_policy.py

@@ -0,0 +1,135 @@
+#!/usr/bin/env python3
+"""Enforce the repository's issue/PR metadata policy in Gitea Actions."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import re
+import urllib.error
+import urllib.request
+from pathlib import Path
+from typing import Any
+
+
+ISSUE_REFERENCE = re.compile(
+    r"(?im)\b(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?|part\s+of|"
+    r"related\s+to|refs?|references)\s+#(\d+)\b"
+)
+TRIAGE_LABEL = "Status/Needs Triage"
+
+
+def deliberate_issue_numbers(title: str, body: str) -> set[int]:
+    """Return same-repository issue numbers referenced intentionally."""
+    return {int(match) for match in ISSUE_REFERENCE.findall(f"{title}\n{body}")}
+
+
+class GiteaApi:
+    """Small API client using the Actions-provided repository token."""
+
+    def __init__(self, api_url: str, repository: str, token: str) -> None:
+        self.base = f"{api_url.rstrip('/')}/repos/{repository}"
+        self.token = token
+
+    def request(self, method: str, path: str, payload: object | None = None) -> Any:
+        data = None if payload is None else json.dumps(payload).encode()
+        request = urllib.request.Request(
+            f"{self.base}{path}",
+            data=data,
+            method=method,
+            headers={
+                "Authorization": f"token {self.token}",
+                "Content-Type": "application/json",
+            },
+        )
+        with urllib.request.urlopen(request, timeout=15) as response:
+            if response.status == 204:
+                return None
+            return json.load(response)
+
+
+def check_pull_request(event: dict[str, Any], api: GiteaApi) -> list[str]:
+    """Return policy violations for a pull_request event."""
+    pull = event["pull_request"]
+    errors: list[str] = []
+    labels = pull.get("labels") or []
+    if labels:
+        errors.append(
+            "PRs must be unlabeled; put tracker metadata on the linked issue "
+            f"(found: {', '.join(label['name'] for label in labels)})."
+        )
+
+    numbers = deliberate_issue_numbers(pull.get("title", ""), pull.get("body", ""))
+    if not numbers:
+        errors.append(
+            "PR must reference an issue with Closes/Fixes/Resolves #N, "
+            "Part of #N, Related to #N, Refs #N, or References #N."
+        )
+        return errors
+
+    real_issues = 0
+    for number in sorted(numbers):
+        try:
+            item = api.request("GET", f"/issues/{number}")
+        except urllib.error.HTTPError as error:
+            if error.code == 404:
+                errors.append(f"Referenced issue #{number} does not exist.")
+                continue
+            raise
+        if item.get("pull_request") is not None:
+            errors.append(f"#{number} is a pull request, not an issue.")
+        else:
+            real_issues += 1
+
+    if not real_issues and not errors:
+        errors.append("PR must reference at least one real issue.")
+    return errors
+
+
+def ensure_issue_label(event: dict[str, Any], api: GiteaApi) -> bool:
+    """Apply the triage label if an issue event leaves the issue unlabeled."""
+    issue = event["issue"]
+    if issue.get("pull_request") is not None or issue.get("labels"):
+        return False
+    labels = api.request("GET", "/labels?limit=100")
+    triage = next((label for label in labels if label["name"] == TRIAGE_LABEL), None)
+    if triage is None:
+        raise RuntimeError(f"repository label {TRIAGE_LABEL!r} does not exist")
+    api.request("POST", f"/issues/{issue['number']}/labels", {"labels": [triage["id"]]})
+    return True
+
+
+def _load_event(path: str) -> dict[str, Any]:
+    return json.loads(Path(path).read_text(encoding="utf-8"))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("command", choices=("check-pr", "label-issue"))
+    parser.add_argument("--event", default=os.environ.get("GITHUB_EVENT_PATH"))
+    args = parser.parse_args()
+    if not args.event:
+        parser.error("--event or GITHUB_EVENT_PATH is required")
+
+    api = GiteaApi(
+        os.environ["GITHUB_API_URL"],
+        os.environ["GITHUB_REPOSITORY"],
+        os.environ["GITHUB_TOKEN"],
+    )
+    event = _load_event(args.event)
+    if args.command == "check-pr":
+        errors = check_pull_request(event, api)
+        if errors:
+            print("\n".join(f"::error::{error}" for error in errors))
+            return 1
+        print("PR tracker policy passed.")
+        return 0
+
+    changed = ensure_issue_label(event, api)
+    print(f"Applied {TRIAGE_LABEL}." if changed else "Issue already has a label.")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

tests/unit/test_cli_dispatch.py

@@ -95,5 +95,60 @@ class TestMainDispatch(unittest.TestCase):
             self.assertEqual(130, main(["x"]))
 
 
+class TestMigrationGate(unittest.TestCase):
+    """The dispatcher's schema-migration gate (cli/__init__.py)."""
+
+    def setUp(self) -> None:
+        # Force the "schema out of date" branch for every test here.
+        patcher = patch.object(StoreManager, "is_migrated", return_value=False)
+        patcher.start()
+        self.addCleanup(patcher.stop)
+        self.addCleanup(StoreManager.reset)
+
+    def test_store_command_blocks_when_stdin_cannot_confirm(self) -> None:
+        # Non-TTY stdin at EOF (as in CI): the [y/N] prompt reads "" and the
+        # command is refused rather than migrating silently.
+        ran: list[bool] = []
+
+        def handler(_rest: list[str]) -> int:
+            ran.append(True)
+            return 0
+
+        with patch.dict(climod.COMMANDS, {"list": handler}), \
+                patch("sys.stdin", io.StringIO("")), \
+                patch("sys.stderr", io.StringIO()):
+            self.assertEqual(1, main(["list"]))
+        self.assertEqual([], ran, "gated command must not dispatch")
+
+    def test_backend_command_skips_gate(self) -> None:
+        # `backend` provisions/probes the host and never opens the store, so
+        # it must run even on an unmigrated DB with unanswerable stdin.
+        ran: list[bool] = []
+
+        def handler(_rest: list[str]) -> int:
+            ran.append(True)
+            return 0
+
+        with patch.dict(climod.COMMANDS, {"backend": handler}), \
+                patch("sys.stdin", io.StringIO("")), \
+                patch("sys.stderr", io.StringIO()):
+            self.assertEqual(0, main(["backend", "status"]))
+        self.assertEqual([True], ran, "exempt command must dispatch")
+
+    def test_store_command_migrates_on_confirmation(self) -> None:
+        migrated: list[bool] = []
+
+        def handler(_rest: list[str]) -> int:
+            return 0
+
+        with patch.dict(climod.COMMANDS, {"list": handler}), \
+                patch.object(StoreManager, "migrate",
+                             side_effect=lambda: migrated.append(True)), \
+                patch("sys.stdin", io.StringIO("y\n")), \
+                patch("sys.stderr", io.StringIO()):
+            self.assertEqual(0, main(["list"]))
+        self.assertEqual([True], migrated, "confirmed gate must migrate")
+
+
 if __name__ == "__main__":
     unittest.main()

tests/unit/test_cli_start_headless.py

@@ -184,6 +184,18 @@ class TestCmdStartHeadless(unittest.TestCase):
         )
         self.assertEqual("docker", self._launch_mock.call_args[1]["backend_name"])
 
+    def test_cached_images_sets_cached_policy(self):
+        start_mod.cmd_start(
+            ["--headless", "--cached-images", "researcher", "--bottle", "claude",
+             "--prompt", "Do it"]
+        )
+        self.assertEqual("cached", self._spec().image_policy)
+
+    def test_cached_images_requires_headless(self):
+        with self.assertRaises(Die):
+            start_mod.cmd_start(["--cached-images", "researcher"])
+        self._launch_mock.assert_not_called()
+
 
 class TestPrepareWithPreflight(unittest.TestCase):
     """prepare_with_preflight calls render_preflight with the plan and backend name."""

tests/unit/test_cli_start_selector.py

@@ -57,6 +57,12 @@ class TestCmdStartSelector(unittest.TestCase):
         self._bottle_picker_mock = self._bottle_picker_patch.start()
         self._bottle_picker_mock.return_value = ["claude"]  # default: one bottle selected
 
+        self._image_policy_patch = patch(
+            "bot_bottle.cli.start._select_image_policy",
+            return_value="fresh",
+        )
+        self._image_policy_patch.start()
+
         self._env_patch = patch.dict(os.environ, {}, clear=False)
         self._env_patch.start()
         os.environ.pop("BOT_BOTTLE_BACKEND", None)
@@ -66,6 +72,7 @@ class TestCmdStartSelector(unittest.TestCase):
         self._launch_patch.stop()
         self._agent_picker_patch.stop()
         self._bottle_picker_patch.stop()
+        self._image_policy_patch.stop()
         self._env_patch.stop()
 
     # ------------------------------------------------------------------
@@ -124,6 +131,19 @@ class TestCmdStartSelector(unittest.TestCase):
         spec = self._launch_mock.call_args[0][0]
         self.assertEqual(("claude", "dev"), spec.bottle_names)
 
+    def test_image_policy_forwarded_to_spec(self):
+        with patch("bot_bottle.cli.start._select_image_policy", return_value="cached"):
+            start_mod.cmd_start(["researcher"])
+        self._launch_mock.assert_called_once()
+        spec = self._launch_mock.call_args[0][0]
+        self.assertEqual("cached", spec.image_policy)
+
+    def test_image_policy_cancel_returns_0(self):
+        with patch("bot_bottle.cli.start._select_image_policy", return_value=None):
+            rc = start_mod.cmd_start(["researcher"])
+        self.assertEqual(0, rc)
+        self._launch_mock.assert_not_called()
+
     def test_empty_bottle_selection_forwarded(self):
         self._bottle_picker_mock.return_value = []
         start_mod.cmd_start(["researcher"])
@@ -215,6 +235,7 @@ class TestCmdStartLabelCollision(unittest.TestCase):
         ).start()
         # Stub the bottle picker to always return a selection.
         patch.object(tui_mod, "filter_multiselect", return_value=["claude"]).start()
+        patch("bot_bottle.cli.start._select_image_policy", return_value="fresh").start()
         self.addCleanup(patch.stopall)
 
     def test_no_collision_proceeds_without_reprompt(self):

tests/unit/test_cli_start_stale.py

@@ -0,0 +1,146 @@
+"""Unit: _launch_bottle StaleImageError handling.
+
+Exercises prelaunch_checks / backend.launch flow:
+  - headless mode → die on stale
+  - interactive mode, user declines → stop without launching
+  - interactive mode, user confirms → skip stale check and launch once
+"""
+
+from __future__ import annotations
+
+import io
+import tempfile
+import unittest
+from types import SimpleNamespace
+from typing import Any, cast
+from unittest.mock import MagicMock, patch
+
+from bot_bottle.image_cache import StaleImageError
+from bot_bottle.log import Die
+
+
+def _fake_plan() -> Any:
+    provision = SimpleNamespace(startup_args=())
+    return cast(Any, SimpleNamespace(
+        agent_provision=provision,
+        agent_provider_template="claude",
+        slug="dev-abc",
+    ))
+
+
+def _ok_cm(bottle: Any) -> MagicMock:
+    """Return a context-manager mock that yields `bottle`."""
+    cm = MagicMock()
+    cm.__enter__ = MagicMock(return_value=bottle)
+    cm.__exit__ = MagicMock(return_value=False)
+    return cm
+
+
+class TestLaunchBottleStaleHandling(unittest.TestCase):
+    def setUp(self) -> None:
+        self._tmp = tempfile.mkdtemp(prefix="cli-stale-test.")
+
+    def _spec(self) -> Any:
+        from bot_bottle.backend import BottleSpec
+        from bot_bottle.manifest import ManifestIndex
+        idx = ManifestIndex.from_json_obj({
+            "bottles": {"dev": {}},
+            "agents": {"demo": {"skills": [], "prompt": "", "bottle": "dev"}},
+        })
+        return BottleSpec(
+            manifest=idx,
+            agent_name="demo",
+            copy_cwd=False,
+            user_cwd=self._tmp,
+            identity="dev-abc",
+        )
+
+    def _run_launch(self, **patch_kwargs: Any) -> int:
+        import bot_bottle.cli.start as start_mod
+        spec = self._spec()
+        with patch.object(start_mod, "prepare_with_preflight",
+                          return_value=(_fake_plan(), "dev-abc")), \
+             patch.object(start_mod, "settle_state"), \
+             patch.object(start_mod, "info"):
+            return start_mod._launch_bottle(
+                spec,
+                dry_run=False,
+                backend_name="docker",
+                **patch_kwargs,
+            )
+
+    def test_headless_stale_calls_die(self) -> None:
+        """In headless mode (assume_yes=True), a StaleImageError from prelaunch_checks must call die()."""
+        import bot_bottle.cli.start as start_mod
+
+        backend_mock = MagicMock()
+        backend_mock.prelaunch_checks.side_effect = StaleImageError("image is 5 day(s) old")
+
+        with patch.object(start_mod, "get_bottle_backend", return_value=backend_mock), \
+             patch.object(start_mod, "die", side_effect=Die()):
+            with self.assertRaises(Die):
+                self._run_launch(assume_yes=True)
+
+        backend_mock.launch.assert_not_called()
+
+    def test_interactive_user_declines_stops_before_launch(self) -> None:
+        """Interactive user answering 'n' → launch is never called."""
+        import bot_bottle.cli.start as start_mod
+
+        backend_mock = MagicMock()
+        backend_mock.prelaunch_checks.side_effect = StaleImageError("image is 5 day(s) old")
+
+        with patch.object(start_mod, "get_bottle_backend", return_value=backend_mock), \
+             patch.object(start_mod, "read_tty_line", return_value="n"), \
+             patch("sys.stderr", new_callable=io.StringIO):
+            rc = self._run_launch(assume_yes=False)
+
+        self.assertEqual(0, rc)
+        backend_mock.launch.assert_not_called()
+
+    def test_interactive_user_confirms_launches_once(self) -> None:
+        """Interactive user answering 'y' → prelaunch stale error is bypassed; launch called once."""
+        import bot_bottle.cli.start as start_mod
+
+        bottle_mock = MagicMock()
+        bottle_mock.name = "dev-abc"
+
+        backend_mock = MagicMock()
+        backend_mock.prelaunch_checks.side_effect = StaleImageError("image is 5 day(s) old")
+        backend_mock.launch.return_value = _ok_cm(bottle_mock)
+
+        with patch.object(start_mod, "get_bottle_backend", return_value=backend_mock), \
+             patch.object(start_mod, "read_tty_line", return_value="y"), \
+             patch.object(start_mod, "attach_agent", return_value=0), \
+             patch.object(start_mod, "capture_claude_session_state"), \
+             patch("sys.stderr", new_callable=io.StringIO):
+            rc = self._run_launch(assume_yes=False)
+
+        self.assertEqual(0, rc)
+        backend_mock.prelaunch_checks.assert_called_once()
+        backend_mock.launch.assert_called_once()
+
+    def test_interactive_yes_uppercase_also_accepted(self) -> None:
+        """'Y' or 'YES' should also be accepted as confirmation."""
+        import bot_bottle.cli.start as start_mod
+
+        bottle_mock = MagicMock()
+        bottle_mock.name = "dev-abc"
+
+        backend_mock = MagicMock()
+        backend_mock.prelaunch_checks.side_effect = StaleImageError("image is 5 day(s) old")
+        backend_mock.launch.return_value = _ok_cm(bottle_mock)
+
+        with patch.object(start_mod, "get_bottle_backend", return_value=backend_mock), \
+             patch.object(start_mod, "read_tty_line", return_value="YES"), \
+             patch.object(start_mod, "attach_agent", return_value=0), \
+             patch.object(start_mod, "capture_claude_session_state"), \
+             patch("sys.stderr", new_callable=io.StringIO):
+            rc = self._run_launch(assume_yes=False)
+
+        self.assertEqual(0, rc)
+        backend_mock.launch.assert_called_once()
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_config_store.py

@@ -0,0 +1,86 @@
+"""Unit tests for the host-side configuration store."""
+
+from __future__ import annotations
+
+import sqlite3
+import tempfile
+import unittest
+from pathlib import Path
+
+from bot_bottle.config_store import (
+    DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS,
+    ConfigStore,
+)
+from bot_bottle.store_manager import StoreManager
+
+
+class TestConfigStore(unittest.TestCase):
+    def test_cached_image_warning_days_defaults_to_one(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            store = ConfigStore(Path(tmp) / "bot-bottle.db")
+            store.migrate()
+            self.assertEqual(
+                DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS,
+                store.cached_image_stale_warning_days(),
+            )
+
+    def test_cached_image_warning_days_reads_value(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            store = ConfigStore(Path(tmp) / "bot-bottle.db")
+            store.migrate()
+            store.set_cached_image_stale_warning_days(7)
+            self.assertEqual(7, store.cached_image_stale_warning_days())
+
+    def test_config_schema_uses_explicit_settings_columns(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            store = ConfigStore(Path(tmp) / "bot-bottle.db")
+            store.migrate()
+            with sqlite3.connect(store.db_path) as conn:
+                conn.row_factory = sqlite3.Row
+                columns = [
+                    row["name"]
+                    for row in conn.execute("PRAGMA table_info(bot_bottle_config)")
+                ]
+            self.assertEqual([
+                "id",
+                "cached_image_stale_warning_days",
+            ], columns)
+
+    def test_store_manager_includes_config_store(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            db = Path(tmp) / "bot-bottle.db"
+            manager = StoreManager(db)
+            self.assertFalse(manager.is_migrated())
+            manager.migrate()
+            self.assertTrue(manager.is_migrated())
+
+    def test_cached_image_warning_days_returns_default_when_db_missing(self) -> None:
+        # When the db file doesn't exist yet (parent exists, file doesn't),
+        # the store returns the default without touching the file.
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            store = ConfigStore(Path(tmp) / "missing.db")
+            self.assertEqual(
+                DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS,
+                store.cached_image_stale_warning_days(),
+            )
+
+    def test_cached_image_warning_days_returns_default_on_null_value(self) -> None:
+        # If the row exists but the value is NULL (or not castable to int),
+        # the store falls back to the default.
+        with tempfile.TemporaryDirectory(prefix="config-store.") as tmp:
+            db_path = Path(tmp) / "bot-bottle.db"
+            store = ConfigStore(db_path)
+            store.migrate()
+            # Write a NULL value directly.
+            with sqlite3.connect(db_path) as conn:
+                conn.execute(
+                    "UPDATE bot_bottle_config SET cached_image_stale_warning_days = NULL WHERE id = 1"
+                )
+            self.assertEqual(
+                DEFAULT_CACHED_IMAGE_STALE_WARNING_DAYS,
+                store.cached_image_stale_warning_days(),
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_docker_launch_committed_image.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import contextlib
+import dataclasses
 import io
 import tempfile
 import unittest
@@ -18,6 +19,7 @@ from bot_bottle.backend.docker.bottle_plan import DockerBottlePlan
 from bot_bottle.backend.docker.consolidated_launch import LaunchContext
 from bot_bottle.egress import EgressPlan
 from bot_bottle.git_gate import GitGatePlan
+from bot_bottle.log import Die
 from bot_bottle.manifest import ManifestIndex
 from tests.unit import use_bottle_root
 
@@ -92,6 +94,7 @@ class TestLaunchCommittedImage(unittest.TestCase):
                 mock.patch.object(launch_mod.docker_mod, "image_exists", return_value=image_present), \
                 mock.patch.object(launch_mod.docker_mod, "build_image", side_effect=_build), \
                 mock.patch.object(launch_mod.docker_mod, "verify_agent_image"), \
+                mock.patch.object(launch_mod.docker_mod, "image_created_at"), \
                 mock.patch.object(launch_mod, "launch_consolidated", return_value=_CTX), \
                 mock.patch.object(launch_mod, "teardown_consolidated"), \
                 mock.patch.object(launch_mod, "DockerGateway", return_value=gw), \
@@ -112,7 +115,8 @@ class TestLaunchCommittedImage(unittest.TestCase):
         with self._patched(
             committed_tag=committed_tag, image_present=image_present, compose=compose,
         ) as built:
-            with launch_mod.launch(plan, provision=mock.Mock(return_value=None)):
+            images = launch_mod.build_or_load_images(plan)
+            with launch_mod.launch(plan, images, provision=mock.Mock(return_value=None)):
                 pass
         return built
 
@@ -127,14 +131,34 @@ class TestLaunchCommittedImage(unittest.TestCase):
             captured.append(p)
             return {"services": {"agent": {}}}
 
-        with self._patched(committed_tag=_COMMITTED_TAG, image_present=True, compose=compose):
-            with launch_mod.launch(_plan(self._tmp), provision=mock.Mock(return_value=None)):
+        with self._patched(committed_tag=_COMMITTED_TAG, image_present=True, compose=compose) as _:
+            plan = _plan(self._tmp)
+            images = launch_mod.build_or_load_images(plan)
+            with launch_mod.launch(plan, images, provision=mock.Mock(return_value=None)):
                 pass
         self.assertEqual(_COMMITTED_TAG, captured[0].image)
 
     def test_falls_back_to_build_when_no_committed_image(self) -> None:
         self.assertEqual([_DEFAULT_IMAGE], self._run_launch(_plan(self._tmp), committed_tag=None))
 
+    def test_cached_images_skip_build_when_present(self) -> None:
+        base = _plan(self._tmp)
+        plan = dataclasses.replace(
+            base,
+            spec=dataclasses.replace(base.spec, image_policy="cached"),
+        )
+        built = self._run_launch(plan, committed_tag=None, image_present=True)
+        self.assertEqual([], built)
+
+    def test_cached_images_die_when_agent_missing(self) -> None:
+        base = _plan(self._tmp)
+        plan = dataclasses.replace(
+            base,
+            spec=dataclasses.replace(base.spec, image_policy="cached"),
+        )
+        with self.assertRaises(Die):
+            self._run_launch(plan, committed_tag=None, image_present=False)
+
     def test_falls_back_to_build_when_committed_image_missing_from_daemon(self) -> None:
         built = self._run_launch(_plan(self._tmp), committed_tag=_COMMITTED_TAG, image_present=False)
         self.assertEqual([_DEFAULT_IMAGE], built)

tests/unit/test_docker_launch_teardown.py

@@ -16,7 +16,7 @@ from pathlib import Path
 from unittest import mock
 
 from bot_bottle.agent_provider import AgentProvisionPlan
-from bot_bottle.backend import BottleSpec
+from bot_bottle.backend import BottleImages, BottleSpec
 from bot_bottle.backend.docker import launch as launch_mod
 from bot_bottle.backend.docker.bottle_plan import DockerBottlePlan
 from bot_bottle.backend.docker.consolidated_launch import LaunchContext
@@ -93,6 +93,8 @@ class TestTeardownWarning(unittest.TestCase):
             orchestrator_url="http://orch:8099",
         )
 
+        images = BottleImages(agent="bot-bottle-claude:latest", sidecar="bot-bottle-sidecars:latest")
+
         with mock.patch.object(launch_mod.docker_mod, "build_image"), \
              mock.patch.object(launch_mod.docker_mod, "verify_agent_image"), \
              mock.patch.object(launch_mod, "launch_consolidated", return_value=ctx), \
@@ -113,7 +115,7 @@ class TestTeardownWarning(unittest.TestCase):
              ), \
              contextlib.redirect_stderr(buf):
             provision = mock.Mock(return_value=None)
-            with launch_mod.launch(plan, provision=provision):
+            with launch_mod.launch(plan, images, provision=provision):
                 pass
 
         output = buf.getvalue()

tests/unit/test_docker_util_image.py

@@ -9,6 +9,7 @@ from __future__ import annotations
 
 import subprocess
 import unittest
+from datetime import timezone
 from unittest.mock import patch
 
 from bot_bottle.backend.docker import util as docker_mod
@@ -26,6 +27,53 @@ def _fail(stderr: str = "boom") -> subprocess.CompletedProcess:  # type: ignore
     )
 
 
+
+class TestImageCreatedAt(unittest.TestCase):
+    def test_parses_docker_timestamp_with_nanoseconds(self):
+        with patch.object(
+            docker_mod.subprocess, "run",
+            return_value=_ok(stdout="2026-07-06T15:33:47.123456789Z\n"),
+        ) as run:
+            created = docker_mod.image_created_at("bot-bottle-claude:latest")
+        self.assertEqual(2026, created.year)
+        self.assertEqual(123456, created.microsecond)
+        self.assertEqual(timezone.utc, created.tzinfo)
+        self.assertEqual(
+            ["docker", "image", "inspect", "--format", "{{.Created}}", "bot-bottle-claude:latest"],
+            run.call_args.args[0],
+        )
+
+    def test_dies_on_inspect_failure(self):
+        with patch.object(
+            docker_mod.subprocess, "run", return_value=_fail("No such image"),
+        ), patch.object(
+            docker_mod, "die", side_effect=SystemExit("die"),
+        ) as die:
+            with self.assertRaises(SystemExit):
+                docker_mod.image_created_at("missing:tag")
+        die.assert_called_once()
+        self.assertIn("missing:tag", die.call_args.args[0])
+
+    def test_dies_on_invalid_timestamp(self):
+        with patch.object(
+            docker_mod.subprocess, "run",
+            return_value=_ok(stdout="not-a-timestamp\n"),
+        ), patch.object(
+            docker_mod, "die", side_effect=SystemExit("die"),
+        ) as die:
+            with self.assertRaises(SystemExit):
+                docker_mod.image_created_at("some:tag")
+        die.assert_called_once()
+        self.assertIn("some:tag", die.call_args.args[0])
+
+    def test_parse_docker_timestamp_no_tzinfo_defaults_to_utc(self):
+        # A bare datetime with no tz offset should be treated as UTC.
+        dt = docker_mod._parse_docker_timestamp("2024-05-01T10:00:00.000000")
+        self.assertIsNotNone(dt.tzinfo)
+        self.assertEqual(timezone.utc, dt.tzinfo)
+
+
+
 class TestCommitContainer(unittest.TestCase):
     def test_runs_docker_commit(self):
         with patch.object(

tests/unit/test_firecracker_image_builder.py

@@ -35,6 +35,15 @@ class TestBuildAgentRootfsDir(unittest.TestCase):
         build.assert_not_called()
         self.assertEqual(base, out)
 
+    def test_cached_lookup_requires_ready_marker(self):
+        digest = image_builder._dockerfile_hash(self.dockerfile)
+        base = self.cache / "rootfs" / f"agent-{digest}"
+        base.mkdir(parents=True)
+        with patch.object(image_builder.util, "cache_dir", return_value=self.cache):
+            self.assertIsNone(image_builder.cached_agent_rootfs_dir(self.dockerfile))
+            (base / ".bb-ready").write_text("ok\n")
+            self.assertEqual(base, image_builder.cached_agent_rootfs_dir(self.dockerfile))
+
     def test_cache_miss_builds_injects_and_marks_ready(self):
         with patch.object(image_builder.util, "cache_dir", return_value=self.cache), \
              patch.object(image_builder, "_build_in_infra") as build, \

tests/unit/test_git_gate.py

@@ -168,16 +168,18 @@ class TestHookRender(unittest.TestCase):
         # Stdin is buffered to a tempfile so both phases can re-read.
         self.assertIn("refs_file=$(mktemp)", hook)
 
-    def test_new_ref_scan_scoped_to_incoming_commits(self):
-        # A new branch (old=all-zeros) must scan only commits new to the
-        # gate, not the full ancestry — otherwise historical findings
-        # block every new-branch push (PRD 0028 / issue #106).
+    def test_scan_scoped_to_incoming_commits(self):
+        # Every non-delete push scans only commits new to the gate, not
+        # the full ancestry and not the `$old..$new` delta — otherwise
+        # historical fixtures block new-branch pushes (PRD 0028 / #106)
+        # and a rebase/force-push onto an advanced main drags in main's
+        # history incl. the sandbox-escape fixtures (#346).
         hook = git_gate_render_hook()
         self.assertIn('log_opts="$new --not --all"', hook)
-        # The old over-broad full-ancestry range must be gone.
+        # Neither the full-ancestry range nor the ancestry-blind delta
+        # range may survive.
         self.assertNotIn('log_opts="$new"', hook)
-        # Existing-branch delta scan is unchanged.
-        self.assertIn('log_opts="$old..$new"', hook)
+        self.assertNotIn('log_opts="$old..$new"', hook)
 
     def test_forward_ssh_is_non_interactive_and_bounded(self):
         # No prompt (BatchMode) and a connect timeout, so an unreachable

tests/unit/test_image_cache.py

@@ -0,0 +1,81 @@
+"""Unit: image_cache.py — check_stale / check_stale_path."""
+
+from __future__ import annotations
+
+import tempfile
+import unittest
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from unittest.mock import patch
+
+from bot_bottle.image_cache import StaleImageError, check_stale, check_stale_path
+
+
+class TestCheckStale(unittest.TestCase):
+    def _run(self, threshold: int, age_days: float) -> None:
+        created = datetime.now(tz=timezone.utc) - timedelta(days=age_days)
+        with patch("bot_bottle.image_cache.ConfigStore") as cs:
+            cs.return_value.cached_image_stale_warning_days.return_value = threshold
+            check_stale("test image", created)
+
+    def test_negative_threshold_never_raises(self):
+        # Threshold < 0 means the check is disabled — always passes.
+        self._run(threshold=-1, age_days=9999)
+
+    def test_zero_threshold_raises_immediately(self):
+        # threshold=0 means any image is stale the moment it exists.
+        with self.assertRaises(StaleImageError):
+            self._run(threshold=0, age_days=0.1)
+
+    def test_within_threshold_does_not_raise(self):
+        # Age well under threshold — should pass silently.
+        self._run(threshold=7, age_days=2)
+
+    def test_at_threshold_does_not_raise(self):
+        # Exactly at the boundary is fine (<=, not <).
+        self._run(threshold=1, age_days=0.9999)
+
+    def test_exceeds_threshold_raises(self):
+        created = datetime.now(tz=timezone.utc) - timedelta(days=3)
+        with patch("bot_bottle.image_cache.ConfigStore") as cs:
+            cs.return_value.cached_image_stale_warning_days.return_value = 1
+            with self.assertRaises(StaleImageError) as ctx:
+                check_stale("agent image 'bot-bottle:latest'", created)
+        self.assertIn("agent image", str(ctx.exception))
+        self.assertIn("day(s) old", str(ctx.exception))
+
+    def test_naive_datetime_treated_as_utc(self):
+        # check_stale calls .astimezone(utc) on the input; naive datetimes
+        # that would be interpreted as local time should still work.
+        # We can't control the local tz in a unit test, so just ensure
+        # no exception is thrown for a very recent naive datetime.
+        naive_now = datetime(2099, 1, 1)  # far future, always "fresh"
+        with patch("bot_bottle.image_cache.ConfigStore") as cs:
+            cs.return_value.cached_image_stale_warning_days.return_value = 1
+            # Should not raise — the image is brand new.
+            check_stale("test image", naive_now)
+
+
+class TestCheckStalePath(unittest.TestCase):
+    def test_delegates_to_check_stale_with_mtime(self):
+        with tempfile.NamedTemporaryFile() as f:
+            path = Path(f.name)
+            with patch("bot_bottle.image_cache.check_stale") as mock_check:
+                check_stale_path("some artifact", path)
+            mock_check.assert_called_once()
+            label, dt = mock_check.call_args.args
+            self.assertEqual("some artifact", label)
+            self.assertIsInstance(dt, datetime)
+            self.assertIsNotNone(dt.tzinfo)
+
+    def test_raises_stale_for_old_file(self):
+        with tempfile.NamedTemporaryFile() as f:
+            path = Path(f.name)
+            with patch("bot_bottle.image_cache.ConfigStore") as cs:
+                cs.return_value.cached_image_stale_warning_days.return_value = 0
+                with self.assertRaises(StaleImageError):
+                    check_stale_path("cached artifact", path)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_macos_container_util.py

@@ -323,5 +323,71 @@ class TestWaitContainerIpv4(unittest.TestCase):
             self.assertEqual("", util.wait_container_ipv4_on_network("c", "net", timeout=-1))
 
 
+class TestMacosContainerImageCreatedAt(unittest.TestCase):
+    def _ok(self, stdout: str) -> "util.subprocess.CompletedProcess":  # type: ignore
+        return util.subprocess.CompletedProcess(
+            args=[], returncode=0, stdout=stdout, stderr="",
+        )
+
+    def _fail(self, stderr: str = "no such image") -> "util.subprocess.CompletedProcess":  # type: ignore
+        return util.subprocess.CompletedProcess(
+            args=[], returncode=1, stdout="", stderr=stderr,
+        )
+
+    def test_parses_iso_timestamp_from_dict(self):
+        payload = '[{"created": "2025-06-01T12:00:00"}]'
+        with patch.object(util.subprocess, "run", return_value=self._ok(payload)):
+            dt = util.image_created_at("bot-bottle-agent:latest")
+        self.assertEqual(2025, dt.year)
+        self.assertEqual(6, dt.month)
+        self.assertEqual(1, dt.day)
+
+    def test_accepts_list_or_dict_input(self):
+        # Container CLI may return a list; we take the first element.
+        payload = '[{"created": "2024-01-15T08:30:00"}]'
+        with patch.object(util.subprocess, "run", return_value=self._ok(payload)):
+            dt = util.image_created_at("some-image:latest")
+        self.assertEqual(2024, dt.year)
+
+    def test_accepts_uppercase_Created_field(self):
+        payload = '[{"Created": "2024-03-20T10:00:00"}]'
+        with patch.object(util.subprocess, "run", return_value=self._ok(payload)):
+            dt = util.image_created_at("some-image:latest")
+        self.assertEqual(2024, dt.year)
+        self.assertEqual(3, dt.month)
+
+    def test_dies_on_nonzero_returncode(self):
+        with patch.object(util.subprocess, "run", return_value=self._fail("not found")), \
+             patch.object(util, "die", side_effect=SystemExit("die")) as die:
+            with self.assertRaises(SystemExit):
+                util.image_created_at("missing:tag")
+        die.assert_called_once()
+        self.assertIn("missing:tag", die.call_args.args[0])
+
+    def test_dies_on_malformed_json(self):
+        with patch.object(util.subprocess, "run", return_value=self._ok("not-json {")), \
+             patch.object(util, "die", side_effect=SystemExit("die")) as die:
+            with self.assertRaises(SystemExit):
+                util.image_created_at("some:tag")
+        die.assert_called_once()
+
+    def test_dies_when_no_created_field(self):
+        payload = '[{"id": "sha256:abc123"}]'
+        with patch.object(util.subprocess, "run", return_value=self._ok(payload)), \
+             patch.object(util, "die", side_effect=SystemExit("die")) as die:
+            with self.assertRaises(SystemExit):
+                util.image_created_at("some:tag")
+        die.assert_called_once()
+        self.assertIn("creation timestamp", die.call_args.args[0])
+
+    def test_dies_on_invalid_timestamp_format(self):
+        payload = '[{"created": "not-a-date"}]'
+        with patch.object(util.subprocess, "run", return_value=self._ok(payload)), \
+             patch.object(util, "die", side_effect=SystemExit("die")) as die:
+            with self.assertRaises(SystemExit):
+                util.image_created_at("some:tag")
+        die.assert_called_once()
+
+
 if __name__ == "__main__":
     unittest.main()

tests/unit/test_stale_checks.py

@@ -0,0 +1,265 @@
+"""Unit: stale-image check functions across backends, and the
+BottleBackend.launch template method (prelaunch_checks + build_or_load_images).
+
+No real images or containers are used — all Docker/container/smolmachine
+calls are mocked at the module boundary."""
+
+from __future__ import annotations
+
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from typing import Any, cast
+from unittest.mock import MagicMock, patch
+
+
+def _bottle_cm(bottle: Any) -> MagicMock:
+    """Return a mock context manager that yields `bottle`."""
+    cm = MagicMock()
+    cm.__enter__ = MagicMock(return_value=bottle)
+    cm.__exit__ = MagicMock(return_value=False)
+    return cm
+
+
+# ---------------------------------------------------------------------------
+# BottleBackend.launch template — prelaunch_checks + _build_or_load_images
+# ---------------------------------------------------------------------------
+
+class TestBottleBackendLaunchTemplate(unittest.TestCase):
+    """Verify the concrete launch() method on BottleBackend calls
+    _build_or_load_images and _launch_impl, and that prelaunch_checks is a no-op
+    on the base class."""
+
+    def _make_backend(self) -> Any:
+        from bot_bottle.backend.docker.backend import DockerBottleBackend
+        return DockerBottleBackend()
+
+    def test_launch_delegates_to_build_or_load_and_launch_impl(self) -> None:
+        from bot_bottle.backend import BottleImages
+        backend = self._make_backend()
+        plan = cast(Any, SimpleNamespace())
+        bottle = MagicMock()
+        images = BottleImages(agent="agent:latest", sidecar="sidecar:latest")
+        with patch.object(
+            backend, "_build_or_load_images", return_value=images,
+        ) as build_mock, patch.object(
+            backend, "_launch_impl",
+            return_value=_bottle_cm(bottle),
+        ) as impl_mock:
+            with backend.launch(plan):
+                pass
+        build_mock.assert_called_once_with(plan)
+        impl_mock.assert_called_once_with(plan, images)
+
+    def test_noop_default_prelaunch_checks(self) -> None:
+        from bot_bottle.backend.docker.backend import DockerBottleBackend
+        from bot_bottle.backend import BottleBackend
+        backend = DockerBottleBackend()
+        # Base-class prelaunch_checks is a no-op — must not raise.
+        BottleBackend.prelaunch_checks(backend, cast(Any, SimpleNamespace()))  # type: ignore[arg-type]
+
+
+# ---------------------------------------------------------------------------
+# Docker backend stale_checks
+# ---------------------------------------------------------------------------
+
+class TestDockerStaleChecks(unittest.TestCase):
+    def _plan(self, policy: str = "cached", slug: str = "dev-abc") -> Any:
+        spec = SimpleNamespace(image_policy=policy)
+        provision = SimpleNamespace(image="bot-bottle-agent:latest")
+        return cast(Any, SimpleNamespace(
+            spec=spec,
+            slug=slug,
+            image="bot-bottle-agent:latest",
+            agent_provision=provision,
+        ))
+
+    def test_fresh_policy_is_noop(self) -> None:
+        from bot_bottle.backend.docker import launch as mod
+        with patch.object(mod, "read_committed_image") as rci, \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan("fresh"))
+        rci.assert_not_called()
+        cs.assert_not_called()
+
+    def test_committed_image_present_checks_only_committed(self) -> None:
+        from bot_bottle.backend.docker import launch as mod
+        from datetime import datetime, timezone
+        ts = datetime(2025, 1, 1, tzinfo=timezone.utc)
+        with patch.object(mod, "read_committed_image", return_value="committed:latest"), \
+             patch.object(mod.docker_mod, "image_exists", return_value=True), \
+             patch.object(mod.docker_mod, "image_created_at", return_value=ts), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan())
+        cs.assert_called_once()
+        self.assertIn("committed:latest", cs.call_args.args[0])
+
+    def test_no_committed_image_checks_agent(self) -> None:
+        from bot_bottle.backend.docker import launch as mod
+        from datetime import datetime, timezone
+        ts = datetime(2025, 1, 1, tzinfo=timezone.utc)
+        plan = self._plan()
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(mod.docker_mod, "image_exists", return_value=True), \
+             patch.object(mod.docker_mod, "image_created_at", return_value=ts), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(plan)
+        cs.assert_called_once()
+        self.assertIn(plan.image, cs.call_args.args[0])
+
+    def test_image_not_present_skips_check(self) -> None:
+        from bot_bottle.backend.docker import launch as mod
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(mod.docker_mod, "image_exists", return_value=False), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan())
+        cs.assert_not_called()
+
+    def test_only_sidecar_missing_checks_only_agent(self) -> None:
+        from bot_bottle.backend.docker import launch as mod
+        from datetime import datetime, timezone
+        ts = datetime(2025, 1, 1, tzinfo=timezone.utc)
+        plan = self._plan()
+
+        def image_exists(ref: str) -> bool:
+            return ref == plan.image  # Only agent present; sidecar missing.
+
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(mod.docker_mod, "image_exists", side_effect=image_exists), \
+             patch.object(mod.docker_mod, "image_created_at", return_value=ts), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(plan)
+        self.assertEqual(1, cs.call_count)
+        self.assertIn(plan.image, cs.call_args.args[0])
+
+    def test_backend_prelaunch_checks_delegates(self) -> None:
+        from bot_bottle.backend.docker.backend import DockerBottleBackend
+        from bot_bottle.backend.docker import launch as mod
+        backend = DockerBottleBackend()
+        plan = self._plan()
+        with patch.object(mod, "stale_checks") as sc:
+            backend.prelaunch_checks(plan)  # type: ignore[arg-type]
+        sc.assert_called_once_with(plan)
+
+
+# ---------------------------------------------------------------------------
+# macOS container backend stale_checks
+# ---------------------------------------------------------------------------
+
+class TestMacosContainerStaleChecks(unittest.TestCase):
+    def _plan(self, policy: str = "cached", slug: str = "dev-abc") -> Any:
+        return cast(Any, SimpleNamespace(
+            spec=SimpleNamespace(image_policy=policy),
+            slug=slug,
+            image="bot-bottle-agent:latest",
+        ))
+
+    def test_fresh_policy_is_noop(self) -> None:
+        from bot_bottle.backend.macos_container import launch as mod
+        with patch.object(mod, "read_committed_image") as rci, \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan("fresh"))
+        rci.assert_not_called()
+        cs.assert_not_called()
+
+    def test_committed_image_present_checks_only_committed(self) -> None:
+        from bot_bottle.backend.macos_container import launch as mod
+        from datetime import datetime, timezone
+        ts = datetime(2025, 1, 1, tzinfo=timezone.utc)
+        with patch.object(mod, "read_committed_image", return_value="committed:latest"), \
+             patch.object(mod.container_mod, "image_exists", return_value=True), \
+             patch.object(mod.container_mod, "image_created_at", return_value=ts), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan())
+        cs.assert_called_once()
+        self.assertIn("committed:latest", cs.call_args.args[0])
+
+    def test_no_committed_image_checks_agent(self) -> None:
+        from bot_bottle.backend.macos_container import launch as mod
+        from datetime import datetime, timezone
+        ts = datetime(2025, 1, 1, tzinfo=timezone.utc)
+        plan = self._plan()
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(mod.container_mod, "image_exists", return_value=True), \
+             patch.object(mod.container_mod, "image_created_at", return_value=ts), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(plan)
+        cs.assert_called_once()
+        self.assertIn(plan.image, cs.call_args.args[0])
+
+    def test_image_not_present_skips_check(self) -> None:
+        from bot_bottle.backend.macos_container import launch as mod
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(mod.container_mod, "image_exists", return_value=False), \
+             patch.object(mod, "check_stale") as cs:
+            mod.stale_checks(self._plan())
+        cs.assert_not_called()
+
+    def test_backend_prelaunch_checks_delegates(self) -> None:
+        from bot_bottle.backend.macos_container.backend import MacosContainerBottleBackend
+        from bot_bottle.backend.macos_container import launch as mod
+        backend = MacosContainerBottleBackend()
+        plan = self._plan()
+        with patch.object(mod, "stale_checks") as sc:
+            backend.prelaunch_checks(plan)  # type: ignore[arg-type]
+        sc.assert_called_once_with(plan)
+
+
+# ---------------------------------------------------------------------------
+# Firecracker cached rootfs selection and stale checks
+# ---------------------------------------------------------------------------
+
+class TestFirecrackerCachedRootfs(unittest.TestCase):
+    def _plan(self, policy: str = "cached") -> Any:
+        return cast(Any, SimpleNamespace(
+            spec=SimpleNamespace(image_policy=policy),
+            slug="dev-abc",
+            image="bot-bottle-agent:latest",
+            dockerfile_path="/repo/Dockerfile",
+            agent_provider_template="claude",
+        ))
+
+    def test_cached_policy_reuses_ready_rootfs_without_building(self) -> None:
+        from bot_bottle.backend.firecracker import launch as mod
+        cached = Path("/cache/rootfs/agent-deadbeef")
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(
+                 mod.image_builder, "cached_agent_rootfs_dir", return_value=cached,
+             ), \
+             patch.object(mod.image_builder, "build_agent_rootfs_dir") as build:
+            self.assertEqual(cached, mod.build_or_load_agent_base(self._plan()))
+        build.assert_not_called()
+
+    def test_cached_policy_fails_when_rootfs_is_missing(self) -> None:
+        from bot_bottle.backend.firecracker import launch as mod
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(
+                 mod.image_builder, "cached_agent_rootfs_dir", return_value=None,
+             ), \
+             patch.object(mod.image_builder, "build_agent_rootfs_dir") as build, \
+             self.assertRaises(SystemExit):
+            mod.build_or_load_agent_base(self._plan())
+        build.assert_not_called()
+
+    def test_stale_checks_use_ready_marker_timestamp(self) -> None:
+        from bot_bottle.backend.firecracker import launch as mod
+        cached = Path("/cache/rootfs/agent-deadbeef")
+        with patch.object(mod, "read_committed_image", return_value=""), \
+             patch.object(
+                 mod.image_builder, "cached_agent_rootfs_dir", return_value=cached,
+             ), \
+             patch.object(mod, "check_stale_path") as check:
+            mod.stale_checks(self._plan())
+        check.assert_called_once_with(f"agent rootfs {cached}", cached / ".bb-ready")
+
+    def test_backend_prelaunch_checks_delegates(self) -> None:
+        from bot_bottle.backend.firecracker.backend import FirecrackerBottleBackend
+        from bot_bottle.backend.firecracker import launch as mod
+        plan = self._plan()
+        with patch.object(mod, "stale_checks") as checks:
+            FirecrackerBottleBackend().prelaunch_checks(plan)
+        checks.assert_called_once_with(plan)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_tracker_policy.py

@@ -0,0 +1,62 @@
+import unittest
+from unittest.mock import Mock
+
+from scripts.tracker_policy import (
+    TRIAGE_LABEL,
+    check_pull_request,
+    deliberate_issue_numbers,
+    ensure_issue_label,
+)
+
+
+class TestDeliberateIssueNumbers(unittest.TestCase):
+    def test_accepts_completing_and_noncompleting_forms(self):
+        self.assertEqual(
+            deliberate_issue_numbers("Fixes #12", "Part of #14; refs #15"),
+            {12, 14, 15},
+        )
+
+    def test_does_not_treat_incidental_number_as_link(self):
+        self.assertEqual(deliberate_issue_numbers("Audit #12", "See PR #14"), set())
+
+
+class TestCheckPullRequest(unittest.TestCase):
+    def test_accepts_unlabelled_pr_linked_to_real_issue(self):
+        api = Mock()
+        api.request.return_value = {"number": 12, "pull_request": None}
+        event = {"pull_request": {"title": "Change", "body": "Part of #12", "labels": []}}
+        self.assertEqual(check_pull_request(event, api), [])
+
+    def test_rejects_labels_and_pr_reference(self):
+        api = Mock()
+        api.request.return_value = {"number": 12, "pull_request": {}}
+        event = {
+            "pull_request": {
+                "title": "Change",
+                "body": "Closes #12",
+                "labels": [{"name": "Kind/Bug"}],
+            }
+        }
+        errors = check_pull_request(event, api)
+        self.assertEqual(len(errors), 2)
+        self.assertIn("unlabeled", errors[0])
+        self.assertIn("not an issue", errors[1])
+
+
+class TestEnsureIssueLabel(unittest.TestCase):
+    def test_adds_triage_label_to_unlabelled_issue(self):
+        api = Mock()
+        api.request.side_effect = [[{"id": 55, "name": TRIAGE_LABEL}], None]
+        event = {"issue": {"number": 405, "labels": [], "pull_request": None}}
+        self.assertTrue(ensure_issue_label(event, api))
+        api.request.assert_any_call("POST", "/issues/405/labels", {"labels": [55]})
+
+    def test_leaves_labelled_issue_unchanged(self):
+        api = Mock()
+        event = {"issue": {"number": 405, "labels": [{"name": "Kind/Documentation"}]}}
+        self.assertFalse(ensure_issue_label(event, api))
+        api.request.assert_not_called()
+
+
+if __name__ == "__main__":
+    unittest.main()