fix: coder.solve() PR bodies now carry the coder's own summary (v0.29.2)

[mabry1985]

Summary

Caught live: portfolio-plugin#38's PR body was literally "[coder.solve rung=greedy gens=1] solved 1-shot" — a diagnostic string, not a description.

Root cause: _WorktreeSolveAdapter.generate() dispatches the coder into its worktree but discards the reply, keeping only the worktree path. Every coder dispatch is promised "your FINAL message becomes the PR description, verbatim" (loop._build_prompt) — but dispatch() had nothing real to report for a coder.solve()-ladder win, so it fell back to its own [coder.solve rung=... gens=...] {note} string. Same failure class as #62 ("coder's final reply becomes the PR body verbatim"), just on a different dispatch path that #62 never touched.

Worse than cosmetic: _verify_goal's NO_TEST_NEEDED escape hatch reads this same text — so a legitimate no-test-needed change (pure refactor, docs-as-code) dispatched through the solve() ladder would always look like a missing test, forcing an unnecessary re-dispatch/escalation cycle.

Fix:

• generate() now captures each candidate's reply (adapter._replies[wt]).
• dispatch() uses the winning candidate's own reply as the result, falling back to the diagnostic string only when there isn't one (a fusion win returns file content, not a natural-language summary — nothing to fall back FROM).

Test plan

• ruff check . / ruff format --check . — clean
• pytest tests/ -q — 281 passed (2 new: winner's reply is used verbatim + the fusion-win fallback still produces the diagnostic string)

🤖 Generated with Claude Code

[protoquinn]

👀 Quinn is reviewing — verdict (PASS / WARN / FAIL) + findings to follow.

[protoquinn]

QA Audit — PR #68 | fix: coder.solve() PR bodies now carry the coder's own summary (v0.29.2)

VERDICT: WARN (CI still queued — re-review once terminal)

---

CI Status

• test: queued

Diff Review

• coder_seam.py: Adds _replies dict to capture each candidate coder's final reply in generate(); dispatch() now uses the winning candidate's own reply as result_text, falling back to the rung/gens diagnostic string for fusion wins. Clean, minimal change — 2 sites touched.
• Version bumps: 0.29.1 → 0.29.2 in protoagent.plugin.yaml and pyproject.toml.
• Tests: Existing test updated to assert the winner's reply is used verbatim; new test_dispatch_falls_back_to_a_diagnostic_string_when_the_winner_has_no_reply covers the fusion-win fallback. Both well-structured.

Observations

• LOW: clawpatch structural review unavailable (500 — repo not in cache). Manual review of the diff found no cross-file issues. The change is scoped entirely within _WorktreeSolveAdapter — no callers modified, no API contract changes.
• The (reply or "").strip() guard in generate() correctly prevents storing empty/whitespace-only replies, ensuring the fallback fires when the coder produces nothing useful.

— Quinn, QA Engineer

[protoquinn]

Submitted COMMENT review on #68.