Run 188 batch: group order, briefs past compaction, review helpers, bug tickets #474

Closed
opened 2026-11-10 11:05:33 -05:00 by cmoriarty · 1 comment
Owner

What happens: run 188 (the re-run after #471) stressed a big change through eight apply groups. Groups ran out of the order their briefs claimed, compacted agents lost their briefs, the code review ran out of context in 5 minutes, and bugs found along the way were dropped. Findings: braid#467, comment 8154.

Fix: the 17 fixes the owner approved one by one (braid#467, comment 8187):

  • Apply groups: the brief names which groups are done and which are not (1); retry a failed group before a new one (2); fixing what your change broke elsewhere is yours (3); each group gets its own tasks, notes and spec (17); a scenario can be "tested by task N" (18).
  • Compaction: the step's brief saved to a file and re-read after a compaction (5); the summary marks guesses (6); the reminder says "the tests you changed" (7).
  • Bugs found along the way go to a file; an end step prepares tickets, filed under auto-approve, else a person edits, approves or dismisses them (8).
  • Code review: per-area reader helpers and less spec text (9); every changed test file named (10). Servers: app takes a command; Braid stops background commands at step end and does not wait on a serving one (11). B11 catches helper reads (12).
  • A seed entry point planned for apps with chance (14); remove the Visual check's browser-check limit (15); page-script limit 180 s to 360 s (19); an agent can cancel a helper (20); a finished step row keeps its compaction count and total time (16).

Done when: the suite and sims pass, no fixture brief grows, and the next full run is read for each.

**What happens:** run 188 (the re-run after #471) stressed a big change through eight apply groups. Groups ran out of the order their briefs claimed, compacted agents lost their briefs, the code review ran out of context in 5 minutes, and bugs found along the way were dropped. Findings: braid#467, comment 8154. **Fix:** the 17 fixes the owner approved one by one (braid#467, comment 8187): - Apply groups: the brief names which groups are done and which are not (1); retry a failed group before a new one (2); fixing what your change broke elsewhere is yours (3); each group gets its own tasks, notes and spec (17); a scenario can be "tested by task N" (18). - Compaction: the step's brief saved to a file and re-read after a compaction (5); the summary marks guesses (6); the reminder says "the tests you changed" (7). - Bugs found along the way go to a file; an end step prepares tickets, filed under auto-approve, else a person edits, approves or dismisses them (8). - Code review: per-area `reader` helpers and less spec text (9); every changed test file named (10). Servers: `app` takes a command; Braid stops background commands at step end and does not wait on a serving one (11). B11 catches helper reads (12). - A seed entry point planned for apps with chance (14); remove the Visual check's browser-check limit (15); page-script limit 180 s to 360 s (19); an agent can cancel a helper (20); a finished step row keeps its compaction count and total time (16). **Done when:** the suite and sims pass, no fixture brief grows, and the next full run is read for each.
Author
Owner

Shipped in PR 476 (d25e101) and PR 477 (f606db1, archive). Production runs f606db1.

What landed: the 17 fixes in comment 8187 of braid#467, as listed in the PR. On run 188's real change the code review's brief is 91,787 characters (it was sent 132,626) and names the main test file among the changed tests; group 5's is 82,228 (was 98,300) and says which groups are done. No fixture brief grew.

Evidence: ./tools/test.sh green; all 87 sims plus the new bugs-found-filed (a noted bug is drafted, auto-approved and filed); the bug gate answered by hand in the console filed its ticket; scripted replays of run 188's code review and group 5 run on the new engine.

Deploy note: d25e101's self-check failed on python-on-path, which read backend/.venv after the finished-run tidy (braid#192) had deleted it, a race that predates this change. PR 477 checks .osf/deps.json instead; f606db1 deployed.

Changes from the plan: the summary's conclusions are marked unconfirmed in the prompt after a compaction (opencode 2 has no compaction-prompt hook); background commands are stopped and their quiet output read through opencode's /api/shell routes, which 2.0.30 has but which only a real run will exercise; any failure there falls back to the 30-minute cap.

Shipped in PR 476 (d25e101) and PR 477 (f606db1, archive). Production runs f606db1. What landed: the 17 fixes in comment 8187 of braid#467, as listed in the PR. On run 188's real change the code review's brief is 91,787 characters (it was sent 132,626) and names the main test file among the changed tests; group 5's is 82,228 (was 98,300) and says which groups are done. No fixture brief grew. Evidence: `./tools/test.sh` green; all 87 sims plus the new `bugs-found-filed` (a noted bug is drafted, auto-approved and filed); the bug gate answered by hand in the console filed its ticket; scripted replays of run 188's code review and group 5 run on the new engine. Deploy note: d25e101's self-check failed on `python-on-path`, which read `backend/.venv` after the finished-run tidy (braid#192) had deleted it, a race that predates this change. PR 477 checks `.osf/deps.json` instead; f606db1 deployed. Changes from the plan: the summary's conclusions are marked unconfirmed in the prompt after a compaction (opencode 2 has no compaction-prompt hook); background commands are stopped and their quiet output read through opencode's `/api/shell` routes, which 2.0.30 has but which only a real run will exercise; any failure there falls back to the 30-minute cap.
Sign in to join this conversation.
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#474
No description provided.