Skip to content

reflect: check for sibling reflect-ci/fix-ci dispatches before finalizing - #4

Closed
EdbertChan wants to merge 1 commit into
mainfrom
reflect-sibling-dispatch-check
Closed

reflect: check for sibling reflect-ci/fix-ci dispatches before finalizing#4
EdbertChan wants to merge 1 commit into
mainfrom
reflect-sibling-dispatch-check

Conversation

@EdbertChan

@EdbertChan EdbertChan commented Aug 17, 2026

Copy link
Copy Markdown
Owner

Summary

  • Found while reflecting on Invoker CI job fleet / a71f331 (12 jobs): 3 separate workflow dispatches independently authored near-identical unmerged fixes to the same file within a 2-hour window, and 4 separate reflect-ci-a71f331-* branches each ran this skill's full lens fan-out against the same commit family.
  • The existing "dispatch-burst" detector (multi-conversation mode) covers this shape, but its "When to invoke" trigger assumes a human asks "why does X keep happening" — an automated reflect-ci-* task has no human to ask, so it never fires in that context.
  • Adds an unconditional git branch --all | grep sibling check to step 1 (locate transcripts) and a matching "When to invoke" bullet naming the automated-dispatch case.

Test plan

  • Read-through: new text is additive, doesn't change existing process steps or examples.
  • N/A — this is a documentation-only skill-prose change with no executable script to run.

Note

Low Risk
Documentation-only change to skills/reflect/SKILL.md with no runtime or CI behavior changes.

Overview
reflect skill prose now covers concurrent Invoker CI repair: several workflows can each open fix-ci-* / reflect-ci-* chains for the same job id, which previously drove duplicate skill edits because dispatch-burst detection only fired when a human asked “why does X keep happening.”

Adds a When to invoke bullet for automated reflect-ci-* tasks: run the sibling check in step 1 without waiting for that question. In step 1 (Locate transcripts), adds Check for sibling reflect passes before finalizing — before spawning reviewers, run git branch --all | grep -E "reflect-ci-<job-id>|fix-ci-<job-id>" and call out siblings in synthesis so the pass knows it may be one of several on the same commit family.

No script or process-step renumbering; additive documentation only.

Reviewed by Cursor Bugbot for commit 5b9305a. Bugbot is set up for automated code reviews on this repo. Configure here.

…zing

Found while reflecting on Invoker CI job fleet/a71f331 (12 jobs): a
single job had 3 separate workflow dispatches each independently
authoring a near-identical unmerged fix to scripts/electron.cjs within
a 2-hour window, and 4 separate reflect-ci-a71f331-* branches each
running this skill's full lens fan-out against the same commit family.
The dispatch-burst detector in multi-conversation mode already covers
this shape, but its trigger assumes a human asks "why does X keep
happening" -- an automated reflect-ci-* task has no human to ask, so
it never fires there. Adds an unconditional git-branch check to step 1
and a matching "When to invoke" bullet for the automated-dispatch case.
EdbertChan added a commit that referenced this pull request Aug 18, 2026
All four predated the a7fc343 refactor (lens table moved to
references/lenses.md) and had gone stale; ported their content onto the
current structure and closing them in favor of this commit:

- PR #1: path/file_path alias normalization in claude-mode dedup, edit-streak,
  and error-by-file detectors (+ both tests). The is_error half was already
  structural on main. 77/77 tests pass.
- PR #2 (folded into #3 by its own description): step-6 unlanded-prior-draft
  check before writing a new lesson.
- PR #3: step-3 sibling-transcript check, History-lens template-identical
  caution (now in references/lenses.md), step-6 commit-as-you-go.
- PR #4: when-to-invoke bullet for automated reflect-ci runs + sibling
  branch grep folded into the step-3 sibling check.
@EdbertChan

Copy link
Copy Markdown
Owner Author

Landed on main in 9b9f97d — all four open PRs were written against the pre-a7fc343 SKILL.md layout and had gone stale (3 of 4 CONFLICTING). This PR's content was ported onto the current structure: see commit message for the per-PR mapping. Tests: 77/77 pass including this PR's two path-alias tests.

@EdbertChan EdbertChan closed this Aug 18, 2026
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