fix(proxy): demote quarantined bridge reattach keys

[Komzpa]

Summary

• route quarantined full-resend HTTP bridge reattaches onto a fresh account-neutral replay key
• clear stale session affinity inputs before session creation so poisoned hard keys cannot rebuild the same wedged bridge
• tighten the quarantine regression test to assert the fresh key and stripped headers

Validation

• uv run --frozen pytest -q tests/unit/test_proxy_http_bridge.py::test_stream_http_bridge_quarantined_full_resend_stays_unanchored_when_reattach_gate_already_false
• uv run --frozen pytest -q tests/unit/test_proxy_http_bridge.py -q

Notes

This is the focused public carrier for the post-compaction live hang class where a quarantined hard session key could suppress the durable anchor but still create/reuse the same poisoned bridge key.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ddbc1fdf3f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fd602a99a0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4123c19249

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 234ccf0bb4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3530210400

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e4820fa56a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

Codex Review: Something went wrong. Try again later by commenting “@codex review”.

Unknown error

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Soju06]

Thanks — the quarantine-demotion core of this PR is in good shape after the review iterations: the generation fence, the projected account-neutral dispatch, the deterministic replay nonce, and the claim→rebind(clear_continuity)→renew advance of the original durable row all trace correctly, and the test coverage for them is thorough. Two blocking items remain in the bundled slimming/namespace scope, plus a scoping request:

1. Orphaned archived-spec rewrite. openspec/changes/archive/2026-08-13-fix-replayed-namespaced-function-call/specs/responses-api-compat/spec.md is rewritten to say replayed tool-call namespaces "MUST be preserved" on live Responses egress, but the final head no longer contains the serializer change that would implement it: app/core/openai/requests.py::to_payload() still strips namespaces by default, the live spec (openspec/specs/responses-api-compat/spec.md:4595) still says "MUST omit", and this PR's own integration test (test_v1_responses_http_bridge_preserves_agent_wait_output_through_wire_normalization) asserts the upstream function_call item has no namespace key. Please revert the archive edit (it's the historical record of what actually landed); if the requirement is genuinely being flipped, that should be its own openspec change with the serializer work.

2. Core WebSocket slimmer not covered. The new live-spec requirement ("Historical response.create slimming MUST preserve outputs for namespaced agent-control calls") is only implemented in the service-facade slimmer. The duplicate _slim_historical_response_input_item in app/core/clients/proxy.py (~L2244), used when the HTTP session bridge is disabled, still replaces collaboration/multi_agent_v1 outputs with the omission notice — the last completed Codex round flagged this and the final push didn't address it (the follow-up review errored out). Either extend the protection there too or scope the spec text to the bridge path.

3. Scope. The title/body describe a focused quarantined-reattach fix, but the PR also carries the agent-control slimming protection (+ two spec files), a _matching_websocket_request_states_for_previous_response_error heuristic change, and test-flake fixes. Could you split the slimming/spec work into its own PR? The quarantine fix looks close to merge-ready on its own, and separating them keeps the spec question from blocking it.

Minor, non-blocking: _AGENT_CONTROL_TOOL_NAMES matches on bare name with no namespace, so a user tool that happens to be called send_input gets its large outputs permanently exempted from slimming and can newly fail size enforcement — consider requiring the namespace for the name-only legacy case or documenting the tradeoff. Also, in the reroute branch the fresh_reattach_anchor_skipped_quarantined event now logs the freshly minted recovery key because bridge_session_key is reassigned before the log call; logging the original quarantined key (with the new key in detail=) would keep the event useful for tracing.

[Soju06]

Rebase nudge: per the 08-16 assessment this is close to mergeable — two scoped blockers plus the post-0816 conflict. Not superseded by anything on main; a rebase + those two fixes gets it over the line.

[coderabbitai]

Warning

Review limit reached

@Komzpa, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 59 minutes

Limit details: You’ve used all 3 included reviews currently available.

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits within each organization.

For paid Pro and Pro+ reviews, CodeRabbit uses a developer's included PR review attempts over the past 7 days to set the current hourly allowance. At typical activity levels, the full plan allowance applies. Higher sustained activity can lower the allowance until earlier attempts leave the 7-day window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5f36ac4b-deb1-40c1-9272-dc660596e325

📥 Commits

Reviewing files that changed from the base of the PR and between f839952 and f38b650.

📒 Files selected for processing (14)

• app/modules/proxy/_service/http_bridge/quarantine.py
• app/modules/proxy/_service/http_bridge/request_submit.py
• app/modules/proxy/_service/http_bridge/service_stubs.py
• app/modules/proxy/_service/http_bridge/streaming.py
• app/modules/proxy/_service/http_bridge/upstream_events.py
• app/modules/proxy/_service/response_create.py
• app/modules/proxy/_service/support.py
• app/modules/proxy/_service/websocket/helpers.py
• openspec/changes/archive/2026-08-13-fix-replayed-namespaced-function-call/specs/responses-api-compat/spec.md
• openspec/specs/responses-api-compat/spec.md
• tests/integration/test_http_responses_bridge.py
• tests/unit/test_durable_bridge_sessions.py
• tests/unit/test_proxy_http_bridge.py
• tests/unit/test_proxy_utils.py

---

_{Comment @coderabbitai help to get the list of available commands.}

[Soju06]

Thanks for the rebase onto beta.2 — CI is fully green and the quarantine-demotion core still traces correctly. However, checking head f38b650, both blocking items from the 08-16 assessment came through the rebase unchanged:

1. The archived-spec rewrite (openspec/changes/archive/2026-08-13-fix-replayed-namespaced-function-call/specs/responses-api-compat/spec.md) is still present. On current main, app/core/openai/requests.py::to_payload() still strips replayed tool-call namespaces by default, and this PR's own integration test (test_v1_responses_http_bridge_preserves_agent_wait_output_through_wire_normalization) asserts the upstream function_call item has no namespace key — so the rewritten archive describes behavior that never shipped. Please revert the archive edit; if the requirement is genuinely being flipped, that needs its own openspec change with the serializer work.

2. The new live-spec MUST ("Historical response.create slimming MUST preserve outputs for namespaced agent-control calls") is still only implemented in the service-facade slimmer. The duplicate _slim_historical_response_input_item in app/core/clients/proxy.py (~L2740), used when the HTTP session bridge is disabled, still replaces collaboration/multi_agent_v1 outputs with the omission notice. Either extend the protection there or scope the spec text to the bridge path.

And the scoping ask stands: the quarantined-reattach fix itself is essentially merge-ready — splitting the slimming/spec work (plus the _matching_websocket_request_states_for_previous_response_error heuristic and flake fixes) into a follow-up PR would let us land the quarantine fix immediately. If splitting is inconvenient, say the word and we can carve out the quarantine core on our side with attribution preserved.

(Also noting the db migration label appears stale — this head touches no migrations/ files — so I'll drop it.)