1# Self-Improving Agent
2 
3A universal self-improvement system that learns from ALL skill experiences and continuously updates the codebase.
4 
5## Overview
6 
7This agent learns from **every skill interaction** to achieve true lifelong learning. It implements a complete feedback loop with multi-memory architecture, self-correction, and evolution markers.
8 
9## Key Features
10 
11- **Multi-Memory Architecture**: Semantic + Episodic + Working memory
12- **Universal Learning**: Learns from ALL skills, not just PRDs
13- **Pattern Extraction**: Converts experiences into reusable patterns
14- **Self-Correction**: Fixes skill guidance when errors occur
15- **Self-Validation**: Periodically verifies skill accuracy
16- **Automatic Updates**: Updates related skills based on learned patterns
17- **Confidence Tracking**: Measures pattern reliability over time
18- **Human-in-the-Loop**: Collects feedback to validate improvements
19 
20## Memory System
21 
22```
23~/.claude/memory/
24├── semantic/       # Patterns, rules, best practices
25├── episodic/       # Specific experiences and episodes
26└── working/        # Current session context
27```
28 
29## How It Works
30 
31```
32Any Skill Completes
33        ↓
34Extract Experience → Identify Patterns → Update Skills → Consolidate Memory
35        ↓                     ↓                  ↓              ↓
36   What happened?    What can we reuse?   Which skills?    Track metrics
37```
38 
39## Installation
40 
41```bash
42ln -s ~/path/to/agent-playbook/skills/self-improving-agent ~/.claude/skills/self-improving-agent
43```
44 
45## Hooks (Optional)
46 
47Wire hooks to capture errors and session-end signals:
48 
49```json
50{
51  "hooks": {
52    "PreToolUse": [
53      {
54        "matcher": "Bash|Write|Edit",
55        "hooks": [
56          { "type": "command", "command": "bash ${SKILLS_DIR}/self-improving-agent/hooks/pre-tool.sh \"$TOOL_NAME\" \"$TOOL_INPUT\"" }
57        ]
58      }
59    ],
60    "PostToolUse": [
61      {
62        "matcher": "Bash",
63        "hooks": [
64          { "type": "command", "command": "bash ${SKILLS_DIR}/self-improving-agent/hooks/post-bash.sh \"$TOOL_OUTPUT\" \"$EXIT_CODE\"" }
65        ]
66      }
67    ],
68    "Stop": [
69      {
70        "matcher": "",
71        "hooks": [
72          { "type": "command", "command": "bash ${SKILLS_DIR}/self-improving-agent/hooks/session-end.sh" }
73        ]
74      }
75    ]
76  }
77}
78```
79 
80## Triggering
81 
82### Automatic
83After ANY skill completes:
84- prd-planner
85- code-reviewer
86- debugger
87- refactoring-specialist
88- etc.
89 
90### Manual
91```
92"自我进化"
93"self-improve"
94"分析今天的经验"
95"总结这次教训"
96```
97 
98## Example Learning
99 
100### Episode
101```yaml
102Skill: debugger
103Situation: Form submission doesn't refresh data
104Root Cause: Empty callback function
105Pattern: Always verify callbacks have implementations
106Confidence: 0.95 → Updates: debugger, prd-implementation-precheck
107```
108 
109### Skill Update
110```markdown
111## Auto-Update (2025-01-11)
112 
113### Pattern Added
114**Callback Verification**: Always verify that callback functions
115passed as props are not empty and actually execute logic.
116 
117**Source**: Episode ep-2025-01-11-003 (3 occurrences)
118**Action**: Added to debugger checklist
119```
120 
121## Research Basis
122 
123- [SimpleMem: Efficient Lifelong Memory](https://arxiv.org/html/2601.02553v1)
124- [ACM Memory Mechanisms Survey](https://dl.acm.org/doi/10.1145/3748302)
125- [Lifelong Learning of LLM Agents](https://arxiv.org/html/2501.07278v1)
126 
127## Templates
128 
129Reusable templates live in `skills/self-improving-agent/templates`:
130- `pattern-template.md`
131- `correction-template.md`
132- `validation-template.md`
133 
134## License
135 
136MIT
137