The run 188 batch: group order, briefs past compaction, bug tickets, review helpers (#474) #476

Merged
cmoriarty merged 8 commits from run188-batch into main 2026-11-10 12:01:51 -05:00
Owner

The 17 fixes the owner approved after run 188 (braid#474; findings on braid#467, comments 8154 and 8187).

  • Apply groups: a group's brief names the groups that are done and those that are not; a failed group is retried before one that has not started; fixing what your own change broke is yours; each group gets its own tasks and the review notes on its files; a scenario can be "tested by task N".
  • After a compaction: each step's brief is kept in the run's tmp folder and the agent is pointed at it; the summary's conclusions are treated as unconfirmed; "send the tests you changed to test".
  • Bugs found along the way: agents note them in .osf/bugs.json; after the PR opens, plan.bugs.draft → human.file-bugs (a sign-off: auto-approve files) → plan.bugs.file.
  • Code review: areas read through reader helpers, requirement names and untested scenarios instead of the spec text, every changed test file named.
  • Background commands stop with their step, and one quiet for 5 minutes is not waited on; an agent can stop a helper; the Visual check has no browser-check limit; page scripts get 360 s.
  • Lint: B11 follows helpers and cut text. Plans give an app with chance a seed. Step row: compactions and total time across attempts.

Verified: ./tools/test.sh green; all 87 sims plus the new bugs-found-filed; no fixture brief grew (code review −26,653, apply group −3,239); on run 188's real change the review brief is 91,787 chars (was 132,626) and group 5's 82,228 (was 98,300); the bug gate answered in the console files the ticket.

Not yet proven on a real run: opencode's shell output and stop calls (2.0.30 has the routes; any failure falls back to today's 30-minute cap), and the stop tool.

🤖 Generated with Claude Code

The 17 fixes the owner approved after run 188 (braid#474; findings on braid#467, comments 8154 and 8187). - **Apply groups:** a group's brief names the groups that are done and those that are not; a failed group is retried before one that has not started; fixing what your own change broke is yours; each group gets its own tasks and the review notes on its files; a scenario can be "tested by task N". - **After a compaction:** each step's brief is kept in the run's tmp folder and the agent is pointed at it; the summary's conclusions are treated as unconfirmed; "send the tests you changed to `test`". - **Bugs found along the way:** agents note them in `.osf/bugs.json`; after the PR opens, `plan.bugs.draft` → `human.file-bugs` (a sign-off: auto-approve files) → `plan.bugs.file`. - **Code review:** areas read through `reader` helpers, requirement names and untested scenarios instead of the spec text, every changed test file named. - **Background commands** stop with their step, and one quiet for 5 minutes is not waited on; an agent can `stop` a helper; the Visual check has no browser-check limit; page scripts get 360 s. - **Lint:** B11 follows helpers and cut text. **Plans** give an app with chance a seed. **Step row:** compactions and total time across attempts. Verified: `./tools/test.sh` green; all 87 sims plus the new `bugs-found-filed`; no fixture brief grew (code review −26,653, apply group −3,239); on run 188's real change the review brief is 91,787 chars (was 132,626) and group 5's 82,228 (was 98,300); the bug gate answered in the console files the ticket. Not yet proven on a real run: opencode's shell output and stop calls (2.0.30 has the routes; any failure falls back to today's 30-minute cap), and the `stop` tool. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid!476
No description provided.