663bc8afa2
Claude Code 2.1.133+ rejects "skills": "./skills" with a "Path escapes plugin directory" warning, even though ./skills resolves to a valid subdirectory inside the plugin root. Per the upstream docs example, the correct value is "skills" (no ./ prefix), which works across all CC versions and other tools (Codex, Gemini CLI, etc.). Applied to all 47 plugin.json files that used "./skills"; updated CLAUDE.md ClawHub publishing constraints to document the new convention and reference the issue. |
||
|---|---|---|
| .. | ||
| .claude-plugin | ||
| agents | ||
| hooks | ||
| reference | ||
| skills | ||
| templates | ||
| CLAUDE.md | ||
| LICENSE | ||
| README.md | ||
| settings.json | ||
Self-Improving Agent
Auto-memory captures. This plugin curates.
A Claude Code plugin that turns auto-memory into a structured self-improvement loop. Analyze what Claude has learned, promote proven patterns to enforced rules, and extract recurring solutions into reusable skills.
Why
Claude Code's auto-memory (v2.1.32+) automatically records project patterns in MEMORY.md. But it has no judgment about what to keep, what to promote, or when entries go stale. This plugin adds the intelligence layer.
The difference:
- MEMORY.md: "I noticed this project uses pnpm" (background note, truncated at 200 lines)
- CLAUDE.md: "Use pnpm, not npm" (enforced instruction, loaded in full)
Promoting a pattern from memory to rules fundamentally changes how Claude treats it.
Commands
| Command | What it does |
|---|---|
/si:review |
Analyze auto-memory — find promotion candidates, stale entries, health metrics |
/si:promote |
Graduate a pattern from MEMORY.md → CLAUDE.md or .claude/rules/ |
/si:extract |
Turn a recurring pattern into a standalone reusable skill |
/si:status |
Memory health dashboard — line counts, capacity, recommendations |
/si:remember |
Explicitly save important knowledge to auto-memory |
Install
Claude Code
/plugin marketplace add alirezarezvani/claude-skills
/plugin install self-improving-agent@claude-code-skills
OpenClaw
clawhub install self-improving-agent
Codex CLI
./scripts/codex-install.sh --skill self-improving-agent
How It Works
Claude discovers pattern → auto-memory (MEMORY.md)
↓
Pattern recurs 2-3x → /si:review flags it
↓
You approve → /si:promote graduates it to CLAUDE.md
↓
Pattern becomes enforced rule, memory entry removed
↓
Space freed for new learnings
What's Included
| Component | Count | Description |
|---|---|---|
| Skills | 5 | review, promote, extract, status, remember |
| Agents | 2 | memory-analyst, skill-extractor |
| Hooks | 1 | PostToolUse error capture (zero overhead on success) |
| Reference docs | 3 | memory architecture, promotion rules, rules directory patterns |
| Templates | 2 | rule template, skill template |
Design Principles
- Don't fight auto-memory — orchestrate it. Auto-memory captures. This plugin curates.
- No duplicate storage. Reads from
~/.claude/projects/directly. No.learnings/directory. - Zero capture overhead. Auto-memory handles capture. Hook only fires on errors.
- Promotion = graduation. Moving a pattern from MEMORY.md to CLAUDE.md changes its priority.
- Respect the 200-line limit. Actively manages MEMORY.md capacity.
Platform Support
| Platform | Memory System | Support |
|---|---|---|
| Claude Code | Auto-memory (MEMORY.md) | ✅ Full |
| OpenClaw | workspace/MEMORY.md | ✅ Adapted |
| Codex CLI | AGENTS.md | ✅ Adapted |
| GitHub Copilot | copilot-instructions.md | ⚠️ Manual |
Credits
Inspired by pskoett/self-improving-agent — a structured learning loop for AI coding agents. This plugin builds on that concept by integrating natively with Claude Code's auto-memory system.
License
MIT — see LICENSE