Skip to content

✨ [feat] collect: prevent recurring mistakes at their boundary - #58

Merged
EricTechPro merged 1 commit into
mainfrom
codex/correction-prevention
Oct 14, 2026
Merged

EricTechPro merged 1 commit into
mainfrom
codex/correction-prevention

Conversation

@EricTechPro

@EricTechPro EricTechPro commented Oct 14, 2026 •

Copy link
Copy Markdown
Owner

Note

⏳ In review · round 1 · ✅ 5/5 AC · head 6eed57b

Problem

Recurring PR findings could become another call-site patch or reminder without naming the boundary that permits the mistake. The calibration brief also needs a clear distinction between proposing stronger product prevention and its existing skill-only edit authority.

Solution

  • Guide recurring intake toward the strongest appropriate correction: remove an invalid path or reuse an adequate owner, then types/schema, actionable guards, behavioral witnesses and judgment prose.
  • Keep Context, Fix and AC sections; require historical proof to fail old behavior for the right reason, pass after correction and survive behavior-preserving refactors.
  • Keep calibration edits inside the explicit existing canonical SKILL.md allowlist. Product changes remain verified, curated Backlog proposals.
  • Exercise the existing public CLI's exact and exceeded receipt, evidence and allowlist limits, with byte-identical ready/skipped inputs and targets.

Acceptance criteria

  • Recurrence guidance proposes prevention at the permitting boundary and preserves existing ticket shape.
    ↳ Standard/template review, 14 writing-format checks and 7 resolved reference links.
  • The canned recurrence fixture retains dedupe and holding-column behavior.
    ↳ 63 collect-filer checks and 21 collect-fetch tests; these prove intake, not a live date regression.
  • Calibration remains bounded and read-only at the public prepare CLI.
    ↳ 14 tests: exact/exceeded 64 KiB receipt, 32 KiB file, 32 unique files, 128 KiB total, 20 canonical targets; ready/skipped files remain unchanged.
  • The full offline safety suite and all required cross-platform CI checks pass on the final head.
    ↳ 57 local safety suites; all 13 jobs green at 6eed57b: https://github.com/EricTechPro/super-board/actions/runs/38518050786
  • Fresh independent review finds no unresolved defects.
    ↳ Fresh reviewer found no defects at 6eed57b against d4a5da4.

Iteration history

Lane Done Time Details
🔨 builder ✅ Oct 13, PDT focused checks pass; draft prepared
🔍 qa ✅ Oct 13, PDT 57 local safety suites; 13 CI checks at 6eed57b
🧐 reviewer ✅ Oct 13, PDT independent review: no findings at 6eed57b

Risk

🟢 Low · Existing intake and calibration authority remain unchanged. Correction selection is agent judgment; offline checks do not prove model quality. The date fixture is canned evidence, with no product implementation here to reproduce. The eval rubric is updated but no paid eval ran.

- Prefer fitting structural prevention and retained behavioral proof\n- Keep calibration proposals inside existing edit authority\n- Verify public CLI bounds and read-only inputs; 57 safety suites pass
@EricTechPro
EricTechPro marked this pull request as ready for review October 14, 2026 05:49
@qodo-code-review

Copy link
Copy Markdown

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

@EricTechPro
EricTechPro merged commit e29c504 into main Oct 14, 2026
13 checks passed
@EricTechPro
EricTechPro deleted the codex/correction-prevention branch October 14, 2026 05:49
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