Skip to content

v2.0.7

Latest

Choose a tag to compare

@docker-agent docker-agent released this 08 Sep 20:06
· 0 commits to main since this release
Immutable release. Only release title and notes can be modified.
bd672e0

What's Changed

Bug fixes

  • fix: "Review incomplete" regression (concurrent transfer_task race) — Fixes the ~19% "Review incomplete" spike introduced in v2.0.6. When the root agent emitted multiple transfer_task calls in one LLM response, docker-agent's parallel tool dispatcher raced on the shared current-agent pointer, causing validation to fail with Agent drafter cannot transfer task to drafter: target agent not in sub-agents list. Two complementary mitigations:

    • parallel_tool_calls: false on the sonnet model config (API-level guarantee — the model never returns two tool calls in one response)
    • "Sequential delegation (REQUIRED)" prompt instruction (defense-in-depth — one transfer_task per response, wait for the result before the next chunk)
    • fix: disable parallel tool calls in pr-review; bump docker-agent to v1.131.0 by @derekmisler in #99
  • fix: false LGTM fallback replaced with incomplete-run notice — When a transfer_task delegation failed, the post-summary defense-in-depth fallback was posting a synthetic "No issues found — LGTM" review, masking a broken pipeline as an approval. Replaced with an explicit ⚠️ Review did not complete notice. The transfer_task failure path in pr-review.yaml is also made unambiguous: on any delegation error, post the incomplete-review COMMENT immediately and stop (no self-reviewing by reading diff files directly).

    • fix(review-pr): replace false LGTM fallback; harden transfer_task failure path by @derekmisler in #98
  • fix: review timeout bumped 2700 → 3600 s — Sequential chunk delegation takes longer than batched on 4-6 chunk PRs. Budget increased from 45 to 60 min; job timeout-minutes 60 → 75; lock TTL 3600 → 4500 s.

Other changes

Full Changelog: v2.0.6...v2.0.7