Skip to content

fix(agents): a stopped run's uncommitted work is kept, and agents stop mistaking a refused docker CLI for no Docker - #339

Merged
hoangsnowy merged 1 commit into
mainfrom
fix/recovery-commits-unfinished-work
Nov 4, 2026
Merged

hoangsnowy merged 1 commit into
mainfrom
fix/recovery-commits-unfinished-work

Conversation

@hoangsnowy

Copy link
Copy Markdown
Owner

Summary

Two lab failures on VNChess (2026-11-04):

Test plan

  • ReplyRecoveryTests (new: wip commit line, never force-add, no redo/extend) + Acceptance / IssueWork prompt and agent tests — 91 passed
  • CI green
  • Lab: the next run stopped at its limit leaves a wip: commit on its branch; no more "Docker blocked" e2e skips

🤖 Generated with Claude Code

https://claude.ai/code/session_018WAdQ7Rm24xss5QDSvXtJB

…p mistaking a refused docker CLI for no Docker

Two lab failures (VNChess, 2026-11-04):
- #144's run hit its time limit with the whole fix uncommitted; the reply-recovery follow-up was told not to commit,
  so the work was lost. The follow-up now commits everything the stopped run left uncommitted as one
  "wip: unfinished work from the stopped run" commit (gitignored secrets stay out, nothing force-added) and pushes,
  so the next round continues from it; review and acceptance still judge it. The Coding prompts also say to commit and
  push after each working step.
- #148's run tried the docker CLI, which the runner refuses on purpose, and concluded the app host and the e2e suite
  could not run. The Coding and Testing prompts now say Docker runs on the computer, the app host starts its own
  containers, and a refused docker CLI says nothing about that.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018WAdQ7Rm24xss5QDSvXtJB
@hoangsnowy
hoangsnowy merged commit 7e4791e into main Nov 4, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant