Skip to content

Commit f061334

Browse files
jleo3claude
andcommitted
Capture failed acceptance criteria on a /work bail-out
The loop caps at three iterations and bails saying the criteria were probably ambiguous — diagnosing its own most common failure and then throwing the finding away. It now writes an entry to ~/diaspar/artifacts/criteria/ before stopping, which is the only moment that knowledge exists. Re-stamped in place from the Diaspar template rather than waiting for a deliberate re-stamp; this is the only repo running the loop today. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 parent f9d8a16 commit f061334

1 file changed

Lines changed: 6 additions & 0 deletions

File tree

.claude/commands/work.md

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -33,4 +33,10 @@ Set it back with `bd update <THE BEAD> --status blocked` and a note saying
3333
what remains contested. Summarize it for me — that usually means the
3434
acceptance criteria were ambiguous, which is my problem to fix, not yours.
3535

36+
**Then write down what failed.** Append to ~/diaspar/artifacts/criteria/ a
37+
short entry naming the criterion that could not be settled, why it was
38+
ambiguous, and the wording that would have been unambiguous. This is the only
39+
moment that knowledge exists — do not skip it because the session is ending.
40+
41+
3642
Never run `bd close`. The bead closes after merge, not after the PR opens.

0 commit comments

Comments
 (0)