[sena-labs]

Summary

Closes the one review finding on #25 that was still open on the current head. Targets feat/openrouter-api-alignment-phase1 rather than main, so it lands inside #25 instead of racing it.

A streamed reasoning block whose signature arrives in a final, unindexed delta was split into two records — the accumulated text, and a separate signature-only entry. Replayed into the next tool round the signature no longer accompanied the text it signs, which is exactly what #25 set out to fix.

Related Issue

Follows up #25 (review thread on openrouter_pipe.py:492). No separate issue.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to break)
• Documentation update
• Refactoring (no functional changes)
• Test improvement

Changes Made

• _accumulate_reasoning reads the seal off the accumulated slot, not the incoming fragment. The condition was detail.get("signature") is not None or detail.get("data") is not None — an incoming fragment carrying a signature always opened a new record. It is now slot.get("signature") is not None: a slot that already holds a signature is complete, so the next fragment starts a new record, while a trailing signature-only delta merges into the block it closes.
• Dropped data from the seal test. The merge loop concatenates data strings, so sealing on data would have split a multi-fragment encrypted payload into one record per fragment — the same defect in a different coat. Keying on the slot's signature leaves those concatenating.
• Three regression tests in test_alignment.py: a trailing signature-only delta stays with its text; a block following a signed one still starts a new record (the separation the old condition was reaching for is preserved); unindexed encrypted fragments concatenate.

The indexed path was already correct — index keys the slot, so the signature merged there. Only streams that omit index were affected.

Before / after

_accumulate_reasoning(s, [{"type": "reasoning.text", "text": "Let me "}])
_accumulate_reasoning(s, [{"type": "reasoning.text", "text": "think."}])
_accumulate_reasoning(s, [{"type": "reasoning.text", "signature": "sig-abc"}])

before: [{"type":"reasoning.text","text":"Let me think."},
         {"type":"reasoning.text","signature":"sig-abc"}]     ← signature stranded
after:  [{"type":"reasoning.text","text":"Let me think.","signature":"sig-abc"}]

Testing

• Assertion suite passes (python test_pipe.py) — 1058 asserts, 0 failed
• Maintenance regressions pass (python -m unittest -v test_maintenance test_alignment) — 60 tests OK (3 new)
• OWUI compatibility smoke — not affected; no change to the save path, frontmatter or valves
• New tests added for the changes
• Integration tests — n/a
• User-visible changes recorded in CHANGELOG.md — no new entry: [Unreleased] already carries "Replay signed/encrypted reasoning and assistant content through native tool rounds" from feat: align chat, media and routing flows with current OpenRouter APIs (Phase 1) #25, and this makes that line true in the unindexed case rather than adding a separate change. Per CONTRIBUTING.md, [Unreleased] holds pending entries only.

Both commands above are exactly what the Unit tests CI job runs.

Screenshots / Demo

N/A — behavioral fix; the before/after above is the reproduction.

Checklist

• My code follows the project's coding standards
• I have added docstrings to new public methods — none added; the three new tests carry docstrings and the changed branch has an explanatory comment
• My changes do not introduce debug print statements beyond [OpenRouter Pipe] logging
• No secrets, API keys, or credentials are included in this PR — changed files scanned
• My commit messages follow Conventional Commits

🤖 Generated with Claude Code

https://claude.ai/code/session_01Rw7cYoHbaXFqdGBTXiRfeh

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-05T19:14:24.475009Z	9ae2073	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.