Agent Self-Improvement
Paste the following prompt into your AI chat to install this skill:
Please follow https://skillhub.cn/install/skillhub.md to install @user_ef64daef/ziwojinhua.
About this skill
Problem It Solves
Long-running agents often hit recurring issues: command failures leave only transient errors, user corrections are not retained, and tool quirks reappear. This skill turns those signals into reviewable Markdown records so later sessions can continue fixing from ERRORS.md, LEARNINGS.md, and FEATURE_REQUESTS.md instead of starting over.
How It Works
The core loop is record first, then promote. When a command fails, a user corrects the agent, knowledge is outdated, or a better approach appears, the skill requires logging to .learnings/ with fields such as ID LRN-20250115-001, Status, Priority, and Area. If a pattern recurs, it links entries with See Also, tracks Recurrence-Count, and promotes distilled rules into CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md. In an OpenClaw workspace, broader behavioral and tool guidance can go into SOUL.md or TOOLS.md.
Boundaries
It fits engineering-oriented agent workflows with project context, writable files, and periodic review. It is not a substitute for root-cause fixes, and it will not make every log valuable. Entries that stay pending for too long or capture one-off details become noise. Use it by keeping records specific, reproducible, and promotable.
Use Cases
- Log Claude Code command failures with reproduction steps and fixes to .learnings/ERRORS.md.
- Record user corrections as LEARNINGS entries with the correction category immediately.
- Aggregate recurring tool gotchas by Pattern-Key and promote rules to AGENTS.md.
- Distill verified cross-project insights into a reusable skill and update the learning entry.
Best For
- Developers maintaining Claude Code or Codex workflows who want corrections to become project memory.
- Agent engineers using OpenClaw multi-session collaboration who need to share tool insights across sessions.
- Engineers owning agent prompt governance who want recurring errors turned into concise prevention rules.
- Internal AI tool maintainers who want verified patterns promoted into reusable skills.
Related Skills
Local workflow memory with matching and SOP updates.
An OpenClaw live streaming executor that initializes TRTC streaming, starts a real-time dashboard, generates viewer URLs, and continuously reports live events.
Breaks down physical supply chains for super-trends to identify second- and third-layer bottlenecks, runs valuation and reverse checks, and maintains trackable reports.
A token-saving compression mode for Chinese LLMs with lite, full, ultra, and classical tiers, preserving code and technical terms while handling edge cases.