fix(responses): recover the null-code encrypted-content rejection and lock pool-switch stripping

[lidge-jun]

Summary

Fixes #2247: ChatGPT account-pool thread continuation could 400 with 'The encrypted content ... could not be verified' when the pool routed a thread to a different account than the one that minted the historical reasoning ciphertext.

• The serving-identity PRE-FLIGHT (from fix(responses): routed-destination sanitize + opaque-state recovery series (#2254 rebased) #2258) was already correct for pool slot switches — now locked by an end-to-end A->B regression proving the second send strips reasoning encrypted_content and compaction blobs, plus a restart-style durable pool-account isolation regression.
• The real gap: the RECOVERY matcher only accepted code: invalid_encrypted_content or the two invalid-argument prefixes. The reported ChatGPT rejection carries invalid_request_error with code: null and the message 'The encrypted content ... could not be verified. Reason: Encrypted content could not be decrypted or parsed.' — added a narrow exact-message matcher for that shape, keeping the fail-closed default.

Verification

• Touched suites 25/0 (267 expects); mutation proof: the strip regression fails with the identity-change flag disabled.
• tsc, privacy:scan green. Full-suite supplemental run green except a known unrelated GUI loader flake.

Checklist

• Targets dev
• Closes Account pool: continuing a thread after pool selection moves to a new account fails with 400 encrypted content could not be verified (account-bound reasoning ciphertext) #2247 evidence in tests

Summary by CodeRabbit

• Bug Fixes

  • Improved recovery when encrypted response content cannot be verified, including ChatGPT-style error responses.
  • Prevented stale encrypted content from being reused after switching between pooled accounts.
  • Ensured failed encrypted-content requests retry once without the unverifiable content.

• Documentation

  • Updated error-handling documentation to cover the additional response format.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 8cb2c1fe-0f46-4cac-85fc-4190f8f699be

📥 Commits

Reviewing files that changed from the base of the PR and between 3ffffc8 and 5d36686.

📒 Files selected for processing (3)

• src/server/responses/core.ts
• tests/responses-opaque-blob-recovery.test.ts
• tests/thought-signature-credential-scope.test.ts

---

📝 Walkthrough

Walkthrough

The response proxy now detects ChatGPT’s code-less unverifiable encrypted-content error. Tests cover pooled-account switching, ciphertext removal before retry, single-retry behavior, and account-scoped replay signatures.

Changes

Opaque blob recovery

Layer / File(s)	Summary
ChatGPT rejection identity src/server/responses/core.ts, tests/responses-opaque-blob-recovery.test.ts	The detector accepts the exact code-less unverifiable encrypted-content message while retaining the existing error-code check. Tests reject unrelated signature errors.
Pooled-account ciphertext recovery tests/responses-opaque-blob-recovery.test.ts, tests/thought-signature-credential-scope.test.ts	Tests verify ciphertext removal when a thread moves between pooled accounts, one retry after ChatGPT rejection, and stable account-specific replay signatures across reloads.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Suggested reviewers: ingwannu

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/pool-ciphertext-regression

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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