AI Agent Hub
Back to skills
Self Reflection Log icon

Self Reflection Log

AI Agent Updated 2026.08.30

Paste the following prompt into your AI chat to install this skill:

Please install @user_2d5dabaa/ziwofansi-ceshi according to https://skillhub.cn/install/skillhub.md.

About this skill

Problem It Solves

AI agents often hit transient failures, user corrections, stale assumptions, and reusable lessons during coding and tool use. Without a persistent record, the same mistake can resurface in the next session. self-improvement moves that context into visible Markdown files instead of relying on ephemeral model memory.

How It Works

The skill defines a lightweight learning log: .learnings/ERRORS.md for command failures, exceptions, and integration errors; .learnings/LEARNINGS.md for corrections, knowledge gaps, and best practices; .learnings/FEATURE_REQUESTS.md for capabilities the user asks for.

The workflow is:
- Create an entry with an ID, Status, Priority, and Area when an error, correction, or feature request appears.
- Link similar entries with See Also and deduplicate recurring patterns using Pattern-Key, Recurrence-Count, and Last-Seen.
- Distill broadly applicable lessons into CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md.
- Promote recurring patterns into short prevention rules when they appear across tasks within a defined window.

Boundaries and Caveats

This fits multi-session or shared-code projects, especially with Claude Code, Codex, and Copilot workflows. It does not perform fixes by itself and does not replace tests or review. Quality depends on logging immediately, capturing reproduction steps, and regularly moving pending entries to resolved or promoted.

Use Cases

  • When fixing repeated failures, log exception output in `.learnings/ERRORS.md` with `Status` and `Priority`.
  • When corrected on project conventions, record the knowledge in `LEARNINGS.md` and later distill it into `CLAUDE.md` or `AGENTS.md`.
  • When a tool failure recurs, deduplicate entries with `Pattern-Key` and increase `Recurrence-Count`.
  • After finishing a feature, review `FEATURE_REQUESTS.md` and `ERRORS.md` and promote reusable rules into project prompts.

Best For

  • Engineers maintaining Claude Code workflows who want to distill corrections and lessons into `CLAUDE.md`.
  • Developers using Copilot for backend code who need to log API behavior differences in `.learnings/LEARNINGS.md`.
  • Tech leads managing multi-agent coding projects who want recurring errors distilled into `AGENTS.md` rules.
  • Engineers debugging Codex toolchain failures who need command failures and reproduction steps in `ERRORS.md`.