Self-Improvement Learning Log
Paste the following prompt into your AI chat to install this skill:
Please install @user_68684d9b/aan according to https://skillhub.cn/install/skillhub.md.
About this skill
Problem
Coding agents keep repeating the same failures in long sessions: command errors, inconsistent API behavior, or knowledge gaps only noticed after user correction. If these stay in chat history, the next session can make the same mistake. This skill turns one-off trial and error into searchable, reviewable, reusable project knowledge.
How It Works
It asks the agent to log at trigger points: command or tool failures go to .learnings/ERRORS.md, user corrections or knowledge gaps go to .learnings/LEARNINGS.md, and missing capabilities go to .learnings/FEATURE_REQUESTS.md. Entries carry Status, Priority, Area, and See Also, making them easier to filter and link. When a lesson applies across files or tasks, it is distilled into CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md; in OpenClaw, it can also go into SOUL.md and TOOLS.md.
The workflow is record first, synthesize second, then make durable. Recurring issues accumulate Recurrence-Count and are promoted into short rules rather than long incident reports. Stable lessons can also be extracted into standalone skills.
Boundaries and Notes
It does not automatically fix code; its core value is an auditable learning loop. Logs must stay specific and avoid presenting guesses as facts. If a team shares .learnings/ in a repo, be careful that entries may contain paths, error output, or other sensitive details.
Use Cases
- After repeated build failures in Claude Code, log command errors to `.learnings/ERRORS.md`.
- When a user flags outdated API usage, record it as a `knowledge_gap` and promote it to `AGENTS.md`.
- After recurring pnpm install issues, distill a reusable rule into `CLAUDE.md` to prevent repeats.
- In a shared repo, triage `.learnings/`, link high-priority issues, and promote durable lessons.
Best For
- Claude Code project engineers who want recurring debugging lessons turned into durable rules.
- Developers using Codex or Copilot who need consistent logs for errors, corrections, and requests.
- Engineers managing multi-agent workflows who want tool gotchas added to agent context.
- AI-assisted engineering leads who need an auditable learning log with promotion rules.
Related Skills
Provides Claw with character-library selection, switching, saving, and global SOUL.md style sync for role-based conversation.
An AIONE Agentic AI Infrastructure SDK wrapper for building production AI agents with memory, skills, workflows, and hooks.
A Python/TypeScript SDK wrapper for the DeepSeek-Reasonix native AI coding agent, with prefix-cache support.
A browser automation tool for analysts, operators, and developers that locates elements, fills forms, extracts structured content, and supports no-code scheduling and export.