feat(supervise+orchestrator): slice 11 — per-bottle supervise queue + DLP safelist

Consolidated egress ran every bottle through one process but keyed the
supervise proposal queue off a single SUPERVISE_BOTTLE_SLUG env and kept
one *global* DLP safelist — so in the shared gateway an operator's token
approval for bottle A would (a) be attributed to the wrong bottle and
(b) leak into bottle B's DLP scan (A's approved secret passes B's egress).
This slice keys both per bottle, resolved by source IP.

- policy_resolver: add `resolve_policy_and_bottle_id` — policy + bottle id
  in one `/resolve`, so egress keys routing *and* the supervise
  queue/safelist from a single round-trip. Fail-closed (403 -> (None,None)).
- egress_addon_core: add `resolve_client_context` (+ `ContextResolverLike`)
  returning `(Config, bottle_id)`, sharing the fail-closed parse with
  `resolve_client_config` via `_config_from_policy`.
- egress_addon: `_active_config` -> `_resolve_flow` returns `(Config, slug)`;
  `safe_tokens` set -> per-bottle `_safe_tokens_for(slug)`; the token-allow
  write/await/archive + the approved-token add all use the resolved slug.
  Single-tenant (no resolver) unchanged — slug = the env SUPERVISE_BOTTLE_SLUG.

New tests cover the resolver, the fail-closed context matrix, and the
cross-tenant isolation (an approval lands only in the calling bottle's
safelist; the proposal is keyed by the source-IP-attributed bottle;
unattributed IPs can't supervise).

Out of scope (noted): the git-gate gitleaks-allow hook + supervise_server
agent-proposal paths, and websocket DLP (still self.config-only, inert in
consolidated mode) — follow-up slices.

pyright 0 errors; pylint 9.83/10; unit suite green (1700 tests; the 13
test_sidecar_init /bin/sleep errors are pre-existing NixOS-local noise).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WBMWTEtQdJ4W5UrWuLHCck

bot_bottle/egress_addon.py

@@ -32,7 +32,7 @@ from egress_addon_core import (  # type: ignore[import-not-found]  # pylint: dis
     is_git_push_request,
     load_config,
     match_route,
-    resolve_client_config,
+    resolve_client_context,
     outbound_scan_headers,
     route_to_yaml_dict,
     scan_inbound,
@@ -99,17 +99,29 @@ class EgressAddon:
         # Absent → single-tenant (static routes file); behaviour unchanged.
         orch_url = os.environ.get(ORCHESTRATOR_URL_ENV, "").strip()
         self._resolver = PolicyResolver(orch_url) if orch_url else None
-        # Tokens the operator has approved this session (PRD 0062). In-memory
-        # only — a restart re-prompts. Mutated only from the asyncio loop that
-        # runs the addon hooks, so no lock is needed.
-        self.safe_tokens: set[str] = set()
+        # Tokens the operator has approved this session (PRD 0062), keyed by
+        # bottle so the shared gateway keeps each bottle's safelist separate —
+        # a global set would let bottle A's approved secret pass bottle B's DLP
+        # scan. In-memory only (a restart re-prompts); mutated only from the
+        # asyncio loop that runs the addon hooks, so no lock is needed.
+        self._safe_tokens: dict[str, set[str]] = {}
         self._supervise_slug = os.environ.get("SUPERVISE_BOTTLE_SLUG", "").strip()
         self._token_allow_timeout = _token_allow_timeout_from_env(os.environ)
         self._reload(initial=True)
         self._install_sighup()
 
-    def _supervise_available(self) -> bool:
-        return bool(self._supervise_slug)
+    @staticmethod
+    def _supervise_available(slug: str) -> bool:
+        """Supervise is reachable for this request iff we resolved a bottle to
+        attribute its proposals to (single-tenant env slug, or a source-IP
+        -attributed bottle id). Empty → fail closed (no queue to write to)."""
+        return bool(slug)
+
+    def _safe_tokens_for(self, slug: str) -> set[str]:
+        """This bottle's operator-approved DLP safelist (PRD 0062), created on
+        first use. Keyed by bottle so the shared gateway never leaks one
+        bottle's approved token into another's scan."""
+        return self._safe_tokens.setdefault(slug, set())
 
     def _reload(self, *, initial: bool = False) -> None:
         try:
@@ -218,19 +230,21 @@ class EgressAddon:
             + "\n"
         )
 
-    def _active_config(self, flow: http.HTTPFlow) -> Config:
-        """The Config to apply to this request. Single-tenant → the static
-        `self.config`. Consolidated → the calling bottle's Config, resolved
-        by source IP (fail-closed to deny-all if unattributed). The identity
-        token, if the agent injected one, is read then stripped so it never
-        leaks upstream."""
+    def _resolve_flow(self, flow: http.HTTPFlow) -> "tuple[Config, str]":
+        """The `(Config, supervise slug)` to apply to this request. Single-tenant
+        → the static `self.config` and the env slug. Consolidated → the calling
+        bottle's Config and its bottle id, resolved by source IP in one
+        round-trip (fail-closed to deny-all + empty slug if unattributed). The
+        identity token, if the agent injected one, is read then stripped so it
+        never leaks upstream. The slug keys both the proposal queue and the
+        per-bottle safelist, so an approval only ever affects its own bottle."""
         if self._resolver is None:
-            return self.config
+            return self.config, self._supervise_slug
         conn = flow.client_conn
         client_ip = conn.peername[0] if conn and conn.peername else ""
         token = flow.request.headers.get(IDENTITY_HEADER, "")
         flow.request.headers.pop(IDENTITY_HEADER, None)
-        return resolve_client_config(self._resolver, client_ip, token)
+        return resolve_client_context(self._resolver, client_ip, token)
 
     async def request(self, flow: http.HTTPFlow) -> None:
         request_path, _, query = flow.request.path.partition("?")
@@ -239,14 +253,14 @@ class EgressAddon:
             self._serve_introspection(flow, request_path)
             return
 
-        config = self._active_config(flow)
+        config, slug = self._resolve_flow(flow)
 
         # DLP outbound scan BEFORE stripping auth — catches tokens the
         # agent tried to smuggle in any header, path, query param, or body.
         # Hostname is included to catch DNS-tunnelling exfiltration attempts.
         route = match_route(config.routes, flow.request.pretty_host)
         if route is not None:
-            if not await self._handle_outbound_dlp(flow, route):
+            if not await self._handle_outbound_dlp(flow, route, slug):
                 return
             # The redact policy may have rewritten the request line; recompute
             # the path/query the git checks below rely on.
@@ -310,6 +324,7 @@ class EgressAddon:
         self,
         flow: http.HTTPFlow,
         route: Route,
+        slug: str,
     ) -> bool:
         """Scan the outbound request and apply the route's on-match policy
         (PRD 0062). Returns True if the request may be forwarded, False if a
@@ -331,7 +346,7 @@ class EgressAddon:
             )
             result = scan_outbound(
                 route, scan_text, os.environ,
-                safe_tokens=self.safe_tokens, crlf_text=crlf_text,
+                safe_tokens=self._safe_tokens_for(slug), crlf_text=crlf_text,
             )
             if result is None or result.severity != "block":
                 return True
@@ -366,10 +381,10 @@ class EgressAddon:
 
             # supervise (default): hold the request for operator approval.
             # Fall back to a hard 403 when supervise isn't wired for the bottle.
-            if not self._supervise_available():
+            if not self._supervise_available(slug):
                 self._block_dlp(flow, result)
                 return False
-            approved = await self._supervise_token_block(flow, request_path, result)
+            approved = await self._supervise_token_block(flow, request_path, result, slug)
             if not approved:
                 return False  # _supervise_token_block wrote the 403 response
             # loop: the approved value is now in safe_tokens; re-scan.
@@ -412,12 +427,16 @@ class EgressAddon:
         flow: http.HTTPFlow,
         request_path: str,
         result: ScanResult,
+        slug: str,
     ) -> bool:
         """Route a token DLP block to the operator's supervisor queue and wait.
 
-        Returns True if the operator approved (the matched value is added to
-        `self.safe_tokens` and the caller re-scans); False if the request must
-        be blocked (a 403 response has been written to `flow`)."""
+        `slug` attributes the proposal to the calling bottle (its own queue +
+        safelist) — in the shared gateway this is what keeps one bottle's
+        approval from unblocking another's request. Returns True if the operator
+        approved (the matched value is added to that bottle's safelist and the
+        caller re-scans); False if the request must be blocked (a 403 response
+        has been written to `flow`)."""
         host = flow.request.pretty_host
         payload = build_token_allow_payload(
             redact_tokens(host, env=os.environ),
@@ -426,7 +445,7 @@ class EgressAddon:
             result,
         )
         proposal = _sv.Proposal.new(
-            bottle_slug=self._supervise_slug,
+            bottle_slug=slug,
             tool=_sv.TOOL_EGRESS_TOKEN_ALLOW,
             proposed_file=payload,
             justification=_TOKEN_ALLOW_JUSTIFICATION,
@@ -449,13 +468,13 @@ class EgressAddon:
             **self._req_ctx(flow),
         }) + "\n")
 
-        response = await self._await_token_response(proposal.id)
-        _sv.archive_proposal(self._supervise_slug, proposal.id)
+        response = await self._await_token_response(proposal.id, slug)
+        _sv.archive_proposal(slug, proposal.id)
 
         if response is not None and response.status in (
             _sv.STATUS_APPROVED, _sv.STATUS_MODIFIED,
         ):
-            self.safe_tokens.add(result.matched)
+            self._safe_tokens_for(slug).add(result.matched)
             if self.config.log >= LOG_BLOCKS:
                 sys.stderr.write(json.dumps({
                     "event": "egress_token_allowed",
@@ -478,6 +497,7 @@ class EgressAddon:
     async def _await_token_response(
         self,
         proposal_id: str,
+        slug: str,
     ) -> "_sv.Response | None":
         """Poll the DB for the operator's response without blocking the
         proxy event loop. Returns the Response, or None on timeout."""
@@ -485,7 +505,7 @@ class EgressAddon:
         deadline = loop.time() + self._token_allow_timeout
         while True:
             try:
-                return _sv.read_response(self._supervise_slug, proposal_id)
+                return _sv.read_response(slug, proposal_id)
             except (OSError, ValueError, KeyError):
                 # Not written yet, or a partial/malformed write — retry until
                 # the deadline, then fail closed.
@@ -539,6 +559,9 @@ class EgressAddon:
         """
         if flow.websocket is None:  # type: ignore[union-attr]
             return
+        # WebSocket DLP runs against the static config only (single-tenant); in
+        # the consolidated gateway self.config has no routes, so this is inert
+        # until websocket routing is made source-IP-aware (a separate slice).
         route = match_route(self.config.routes, flow.request.pretty_host)
         if route is None:
             return
@@ -549,7 +572,7 @@ class EgressAddon:
             # not an injection vector here — scan only for credential leakage.
             result = scan_outbound(
                 route, content, os.environ,
-                safe_tokens=self.safe_tokens, crlf_text="",
+                safe_tokens=self._safe_tokens_for(self._supervise_slug), crlf_text="",
             )
             if result is not None and result.severity == "block":
                 sys.stderr.write(f"egress DLP: {result.reason}\n")

bot_bottle/egress_addon_core.py

@@ -421,6 +421,18 @@ class PolicyResolverLike(typing.Protocol):
         ...
 
 
+def _config_from_policy(policy: "str | None") -> "Config":
+    """Parse a resolved policy blob into a Config, fail-closed: None / empty /
+    unparseable all become a deny-all Config (no routes → every request
+    blocked)."""
+    if not policy:
+        return Config(routes=())  # unattributed or empty → deny-all
+    try:
+        return load_config(policy)
+    except ValueError:
+        return Config(routes=())  # unparseable policy → deny
+
+
 def resolve_client_config(
     resolver: PolicyResolverLike, client_ip: str, identity_token: str = ""
 ) -> "Config":
@@ -433,12 +445,33 @@ def resolve_client_config(
         policy = resolver.resolve(client_ip, identity_token)
     except Exception:  # noqa: BLE001  # pylint: disable=broad-exception-caught
         return Config(routes=())  # orchestrator unreachable/errored → deny
-    if not policy:
-        return Config(routes=())  # unattributed or empty → deny-all
+    return _config_from_policy(policy)
+
+
+class ContextResolverLike(typing.Protocol):
+    """The bit of `policy_resolver.PolicyResolver` `resolve_client_context`
+    needs — one round-trip returning both policy and bottle id."""
+
+    def resolve_policy_and_bottle_id(
+        self, source_ip: str, identity_token: str = ...,
+    ) -> "tuple[str | None, str | None]":
+        ...
+
+
+def resolve_client_context(
+    resolver: ContextResolverLike, client_ip: str, identity_token: str = "",
+) -> "tuple[Config, str]":
+    """The calling client's `(Config, bottle_id)` in one round-trip —
+    **fail-closed**. The Config follows `resolve_client_config`'s deny-all
+    rules; the bottle id is `""` whenever unattributed or the orchestrator
+    errored, which the caller treats as "supervise unavailable for this
+    bottle" (never another bottle's queue). One `/resolve` keys both the
+    per-request egress policy and the per-bottle supervise queue + safelist."""
     try:
-        return load_config(policy)
-    except ValueError:
-        return Config(routes=())  # unparseable policy → deny
+        policy, bottle_id = resolver.resolve_policy_and_bottle_id(client_ip, identity_token)
+    except Exception:  # noqa: BLE001  # pylint: disable=broad-exception-caught
+        return Config(routes=()), ""  # orchestrator unreachable/errored → deny
+    return _config_from_policy(policy), (bottle_id or "")
 
 
 # ---------------------------------------------------------------------------
@@ -833,7 +866,9 @@ __all__ = [
     "is_git_fetch_request",
     "load_config",
     "resolve_client_config",
+    "resolve_client_context",
     "PolicyResolverLike",
+    "ContextResolverLike",
     "match_route",
     "outbound_scan_headers",
     "parse_config",

bot_bottle/policy_resolver.py

@@ -100,5 +100,24 @@ class PolicyResolver:
         bottle_id = payload.get("bottle_id")
         return bottle_id if isinstance(bottle_id, str) and bottle_id else None
 
+    def resolve_policy_and_bottle_id(
+        self, source_ip: str, identity_token: str = "",
+    ) -> tuple[str | None, str | None]:
+        """Both the policy blob and the bottle id in a single `/resolve` — so
+        a caller that needs each (the egress addon: policy for routing, bottle
+        id to key the calling bottle's supervise queue + safelist) makes one
+        round-trip, not two. Returns `(None, None)` when unattributed (a clean
+        `403`). Raises `PolicyResolveError` if the orchestrator can't be
+        reached, so the caller still fails closed."""
+        payload = self._post_resolve(source_ip, identity_token)
+        if payload is None:
+            return None, None
+        policy = payload.get("policy")
+        bottle_id = payload.get("bottle_id")
+        return (
+            policy if isinstance(policy, str) else "",
+            bottle_id if isinstance(bottle_id, str) and bottle_id else None,
+        )
+
 
 __all__ = ["PolicyResolver", "PolicyResolveError"]