Self-Improvement Learning Log
Paste the following prompt into your AI chat to install this skill:
Please follow https://skillhub.cn/install/skillhub.md to install @user_60b84151/test-test-test into your AI assistant.
About this skill
Problem
Coding agents can repeat the same mistakes across long tasks: a failed command leaves no context, a user corrects an approach, and the next session still follows the old assumption. test-etst-test-tes turns these moments into searchable Markdown logs instead of relying on session memory.
How It Works
The skill operates around a .learnings/ directory. When an agent hits an error, correction, knowledge gap, or feature request, it writes to ERRORS.md, LEARNINGS.md, or FEATURE_REQUESTS.md, using IDs like LRN-YYYYMMDD-XXX. The core loop includes:
- Capture triggers: non-zero exit codes, exceptions, “it should be like this” corrections, or API behavior that conflicts with docs.
- State updates: entries move from pending to in_progress, resolved, wont_fix, or promoted.
- Memory promotion: rules that apply across files, prevent repeats, or encode project conventions are distilled into CLAUDE.md, AGENTS.md, .github/copilot-instructions.md, or OpenClaw files such as SOUL.md and TOOLS.md.
- Recurring patterns: similar issues are linked with See Also, prioritized when repeated, and can ingest simplify-and-harden candidate patterns.
Boundaries
It does not auto-fix code or decide which rules enter the repository; value depends on timely logging and periodic review. Teams should choose whether .learnings/ is tracked for shared use or kept local with .gitignore.
Use Cases
- After repeated command failures in Claude Code, record exit codes, reproduction steps, and file paths in `.learnings/ERRORS.md`.
- After a user corrects the package manager to `pnpm`, log the reason in `LEARNINGS.md` and promote it to `CLAUDE.md`.
- In an OpenClaw workspace, collect tool gotchas such as Git auth or API timeouts into `TOOLS.md` for later sessions.
- Dedupe recurring refactor suggestions with `Pattern-Key`, then promote stable rules to `AGENTS.md` after repeated hits.
Best For
- Engineers using Claude Code who want to turn errors and corrections into project rules
- Automation owners maintaining multi-agent workspaces and writing tool gotchas into `SOUL.md` or `TOOLS.md`
- QA engineers reviewing AI coding failures and tracking exit codes or exceptions as actionable entries
- Tech leads extracting recurring debugging lessons into reusable skill templates
Related Skills
A one-shot coding agent built on Claude Code CLI that runs non-interactively, supports a specified workdir, and can be monitored in the foreground or background.
Preview and confirm file sorting by extension, with recursive cleanup, ignore rules, and transactional rollback.
An engineering assistant for static HTML/CSS/JS pages, design-token extraction, IE8-compatible review, and structured delivery.
An engineering workflow for requirement analysis, scenario modeling, risk planning, quality gates, testing, and knowledge capture, with lightweight, standard, and full modes.