skills/evolution/SKILL.md
Memory-driven self-evolution layer for Agent Skills. Captures usage patterns, learns from errors, and automatically generates reusable knowledge. Use when you want your agent to learn from experience, track skill effectiveness, or generate improvement suggestions.
npx skillsauth add ZhanlinCui/Auto-Evolution-Agent-Skills auto-evolutionInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Transform your agent from a static assistant into a learning system that evolves with every interaction.
Traditional Skills: User → Agent → Output (static)
Auto-Evolution: User → Agent → Output
↓
[Capture] → [Memory] → [Learn] → [Evolve]
↓
Agent gets smarter over time
Add to your .claude/settings.json:
{
"hooks": {
"PreToolUse": [{"matcher": "Read|Write|Edit|Bash", "hooks": [{"type": "command", "command": "bash .claude/skills/evolution/hooks/capture.sh \"$TOOL_NAME\" \"$TOOL_INPUT\""}]}],
"PostToolUse": [{"matcher": "Bash", "hooks": [{"type": "command", "command": "bash .claude/skills/evolution/hooks/capture.sh post-bash \"$TOOL_OUTPUT\" \"$EXIT_CODE\""}]}],
"Stop": [{"matcher": "", "hooks": [{"type": "command", "command": "bash .claude/skills/evolution/hooks/reflect.sh"}]}]
}
}
Just use Claude Code as you always do. The system silently captures:
At session end, open reports/dashboard.html for a visual summary.
| Layer | Purpose | Retention | Example | |-------|---------|-----------|---------| | Episodic | Raw events | 7 days | "Used layout.md at 14:32" | | Semantic | Patterns | 30 days | "TypeScript errors often follow X" | | Procedural | How-to knowledge | Permanent | Ready-to-use skill drafts |
┌─────────────────────────────────────────────────────────┐
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Capture │ ──▶ │ Analyze │ ──▶ │ Evolve │ │
│ │ Events │ │ Patterns│ │ Skills │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ ▲ │ │
│ │ ┌──────────┐ │ │
│ └────────── │ Memory │ ◀─────────┘ │
│ │ System │ │
│ └──────────┘ │
│ │
└─────────────────────────────────────────────────────────┘
/retrospective - Session ReviewAsk Claude to generate a session retrospective:
User: /retrospective
Produces:
/evolve - Promote KnowledgeWhen you discover a reusable pattern:
User: This error handling approach is useful. /evolve it to a skill.
Claude will:
community//dashboard - Visual ReportUser: /dashboard
Opens the visual evolution dashboard showing:
Edit config.json:
{
"memory": {
"episodic_ttl_days": 7,
"semantic_ttl_days": 30,
"pattern_threshold": 3
},
"capture": {
"ignore_exit_codes": [1],
"ignore_patterns": ["grep|rg|test"],
"force_patterns": ["error|fatal|panic"]
},
"evolution": {
"auto_draft": true,
"require_validation": true,
"min_occurrences": 3
}
}
evolution/
├── SKILL.md # This file
├── config.json # All settings
├── memory/
│ ├── episodes.jsonl # Raw events (append-only)
│ ├── patterns.json # Detected patterns
│ └── drafts/ # Skill candidates
├── hooks/
│ ├── lib.sh # Shared utilities
│ ├── capture.sh # Event capture
│ └── reflect.sh # Session reflection
├── reports/
│ ├── dashboard.html # Visual dashboard
│ └── sessions/ # Session reports
├── templates/
│ └── skill.md # Skill template
└── community/
└── README.md # Contribution guide
Skills evolve through stages:
[Detected] ──▶ [Draft] ──▶ [Validated] ──▶ [Promoted]
│ │ │ │
│ │ │ └─ Becomes official skill
│ │ └─ Verified to work
│ └─ Structured documentation
└─ Pattern appears 3+ times
Works with any skills library. Just copy evolution/ to your .claude/skills/ directory.
For advanced integration, see ARCHITECTURE.md.
data-ai
Example TaskFlow authoring pattern for inbox triage. Use when messages need different treatment based on intent, with some routes notifying immediately, some waiting on outside answers, and others rolling into a later summary.
data-ai
Example TaskFlow authoring pattern for inbox triage. Use when messages need different treatment based on intent, with some routes notifying immediately, some waiting on outside answers, and others rolling into a later summary.
data-ai
OpenProse VM skill pack. Activate on any `prose` command, .prose files, or OpenProse mentions; orchestrates multi-agent workflows.
data-ai
OpenProse VM skill pack. Activate on any `prose` command, .prose files, or OpenProse mentions; orchestrates multi-agent workflows.