Summary

This tracking issue collects OpenAI Responses-compatible provider regressions with GitHub Actions-backed reproduction evidence related to #89728 or adjacent Responses payload/state/stream normalization boundaries.

This issue does not claim that every cited issue shares the same root cause. It records what is reproduced, what is only narrowly reproduced, and what was removed after repeat retesting.

Root issue

• Custom OpenAI Responses-compatible providers still replay item ids when store support is stripped #89728 — Custom OpenAI Responses-compatible providers still replay item ids when store support is stripped

Root condition:

explicitStore === undefined
shouldStripStore === true

Current replay gating allows replay when payloadPolicy.explicitStore !== false, even when effective store support is stripped.

GitHub Actions evidence

Fixture/source baseline

• Run: https://github.com/snowzlm/openclaw/actions/runs/27005369159
• Commit: 7b6b118ad2b2027934800c8e8020a0ced8a235b2
• Conclusion: success

Direct OpenClaw code-path reproduction

• Run: https://github.com/snowzlm/openclaw/actions/runs/27007044935
• Commit: a5d6e76d65df07740e177d3b7ebde5e9f8efc5df
• Conclusion: success

Strict repeat retest for removal

• Run: https://github.com/snowzlm/openclaw/actions/runs/27012355174
• Commit: 6d77a026126831b1165d28e1457308d796aa6f95
• Conclusion: success

Gap closure retest

• Run: https://github.com/snowzlm/openclaw/actions/runs/27012993404
• Commit: 6b1d026a4366d024261fbfca0eb2faeb1d09eee6
• Conclusion: success

Formal reproduction matrix

Removed / not formal citations

#90094 — content:null strict-provider rejection

Removed from the formal matrix after repeat retesting.

Retest run: https://github.com/snowzlm/openclaw/actions/runs/27012355174

Attempts that did not reproduce content:null:

• builder-empty-tool-result
• builder-text-tool-result
• shared-convert-empty-tool-result
• builder-malformed-user-null-content — throws before provider payload serialization
• builder-malformed-assistant-null-content

Keep #90094 only as watchlist until the exact session-history shape or another transport path can reproduce input[n].content === null in GitHub Actions.

Explicit non-claims

• Slack continuation can lose prior plan context and Responses continuation fails with thinking_signature_invalid #84002: this issue does not claim the Slack follow-up was routed into an empty/main session. The Slack routing probe resolved a thread-specific session key, so the empty-session description is removed from this tracking issue.
• GPT-5.5 reasoning item schema error causes duplicate Discord messages via retry cascade #84484: this issue does not claim duplicate Discord visible sends were reproduced. Only the provider schema precondition is reproduced. A separate delivery/idempotency reproduction is required before claiming duplicate-send closure.
• openai-responses transport sends null content, rejected by strict providers (400 schema error) #90094: this issue does not claim direct code-path reproduction for content:null.
• These issues should not be closed together unless their own fix PRs include matching regression tests.

Suggested first fix lane

Start with #89728 because it has the strongest same-root evidence:

const policyAllowsReplayIds =
  payloadPolicy.explicitStore !== false && !payloadPolicy.shouldStripStore;

Regression coverage should assert:

• storeless custom openai-responses providers do not replay prior Responses item ids;
• protocol-required call_id pairing remains intact for function calls and function call outputs;
• duplicate Responses payload construction paths cannot regress independently.

Closure criteria

For each formal citation:

1. keep or improve the Actions reproduction;
2. link the exact code path;
3. preserve the stated scope limit;
4. split any unreproduced channel-level symptom into a separate issue unless a future Actions test reproduces it.