s-hiraoku/codex-harnesses · Archived

retrospective-codify

Convert review feedback, CI failures, bug fixes, or task retrospectives into a durable project feedback ledger, then make future implementation work consult that ledger before planning.

First seen Aug 22, 2026

Installation

$ npx skills add s-hiraoku/codex-harnesses --skill retrospective-codify

Stronger alternatives

This repository is archived — consider an actively maintained alternative.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from s-hiraoku/codex-harnesses · top by installs.

npx skills add s-hiraoku/codex-harnesses

Browse all from s-hiraoku/codex-harnesses

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

License LICENSE
Default branch main
Open issues 2
Status Archived

Skill metadata

Parsed from SKILL.md frontmatter.

Declared agents codex

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 3,049 B
  • docs SUMMARY.md 213 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

Retrospective Codify

Use this workflow when a task produced feedback that should change how future implementation starts. The goal is not to archive every comment; it is to extract reusable implementation rules from concrete review findings, CI failures, regressions, or user corrections.

Workflow

  1. Identify the feedback source.

- Use PR review comments, CodeRabbit or Codex findings, CI failure analysis, bug reports, user corrections, or a completed task retrospective. - Separate one-off requests from reusable failure patterns.

  1. Read the project feedback ledger.

- Prefer ledger/review-feedback.md when present. - If the project uses another durable task-memory path, use that path and name it in the final report. - If no feedback ledger exists and the pattern is worth keeping, create a compact one.

  1. Extract reusable lessons.

- For each candidate, write the specific issue, root cause, general implementation rule, evidence, and verification signal. - Keep rules operational: future agents should know what to check before coding, not just what went wrong. - Do not store secrets, long pasted comments, private user data, or full logs.

  1. Deduplicate before appending.

- If an existing entry covers the same class of failure, update Last seen, Evidence, or Implementation rule instead of adding a parallel entry. - If the same pattern recurs, strengthen the rule so it is harder to miss during planning.

  1. Feed the ledger into future work.

- When starting implementation, read active feedback-ledger entries before planning. - Carry relevant active rules into the plan, tests, and self-review checklist. - If no entries apply, say that the ledger was checked and no relevant prior feedback was found.

  1. Close or retire stale entries.

- Mark an entry Retired only when project structure, tooling, or policy makes the rule obsolete. - Do not delete history unless it contains sensitive content that should not have been recorded.

Entry Format

Use this compact Markdown shape:

### YYYY-MM-DD - Short Pattern Name

- Status: Active
- Last seen: YYYY-MM-DD
- Source: PR review, CI failure, user correction, bug report, or retrospective
- Trigger: when this kind of implementation work starts
- Issue: what went wrong in the concrete task
- Cause: why the original implementation missed it
- Implementation rule: what future work must check or do before coding
- Evidence: concise links, file paths, PR numbers, or command names
- Verification: check that proves the rule was applied

Final Report

Include:

  • feedback sources reviewed
  • ledger path updated or consulted
  • entries added, updated, skipped as one-off, or retired
  • implementation rules future work should apply
  • any sensitive details intentionally omitted