Self Improving Agent

Enables continuous self-improvement through learning from failures, user corrections, and capability gaps. Integrates with QAVR for learned memory ranking.

kimasplund d60db35 4 files · 25.9 KB Updated Sep 10, 2026

Install

Press C to copy

$ npx skillmds@latest add kimasplund/self-improving-agent

Installs into ~/.claude/skills (or .claude/skills inside a project). Restart Claude Code to pick it up.

＋ Add to collection

README badge

[![SkillMD](https://skillmd.com/badge/kimasplund/self-improving-agent.svg)](https://skillmd.com/skills/kimasplund/self-improving-agent)

Trust & safety

Audits pending

Not yet audited · commit d60db35

Permissions this skill uses

Executes scripts
The instructions run shell commands or bundled script files.

Works with: Claude Code Claude.ai OpenAI Codex +

File contents

Self-Improving Agent

Captures learnings, errors, and corrections to enable continuous improvement across sessions.

Activation Triggers

This skill automatically activates when:

Trigger	What Gets Logged	Location
Command fails	Error type, context, recovery suggestion	`logs/failures_detailed.jsonl`
User corrects	"No, that's wrong...", "Actually..."	`logs/corrections.jsonl`
Missing capability	"Can you X?" where X isn't available	`logs/missing_capabilities.jsonl`
API/tool fails	Failure pattern, suggested fix	`logs/failures_detailed.jsonl`
Better approach found	Optimization learned	`logs/learnings.jsonl`

How It Works

1. Automatic Logging (via hooks)

PostToolUse → enhanced-failure-logger.js → logs failures with context
UserMessage → correction-detector.js → detects "wrong/actually/try again"
Response → capability-tracker.js → detects unfulfilled requests

2. Learning Aggregation

All learnings flow to logs/learnings.jsonl:

{
  "timestamp": "2026-01-26T12:00:00Z",
  "type": "user_correction|tool_failure|missing_capability",
  "category": "factual_error|command_failed|web_browsing",
  "description": "what was learned",
  "source": "which detector"
}

3. Session Start Review

On each session start, recent learnings are shown:

=== Learning Review ===
[Learnings] 47 total entries
[Recent]
  • [tool_failure] Bash failed: timeout - WebFetch to external API...
  • [user_correction] User corrected: "No, use the other file..."
[Corrections] 12 user corrections logged
[Capability Gaps] Top requested:
  • send emails (5x)
  • browse web (3x)

4. QAVR Integration

Successful learnings boost Q-values for related memories, improving future retrieval.

Manual Commands

Review Learnings

# Show all learnings
cat ~/.claude/logs/learnings.jsonl | tail -20

# Show corrections only
cat ~/.claude/logs/corrections.jsonl | jq -s 'group_by(.correction_type) | map({type: .[0].correction_type, count: length})'

# Show capability gaps report
node ~/.claude/scripts/hooks/capability-tracker.js --report

Test Detection

# Test correction detector
node ~/.claude/scripts/hooks/correction-detector.js

# Test failure logger
node ~/.claude/scripts/hooks/enhanced-failure-logger.js

# Test capability tracker
node ~/.claude/scripts/hooks/capability-tracker.js

Learning Categories

Correction Types

factual_error - Wrong information provided
retry_request - User asked to try again
misunderstanding - Misinterpreted the request
failed_solution - Solution didn't work

Failure Types

permission_error - Access denied
not_found - File/resource missing
timeout - Operation timed out
network_error - Connection issues
syntax_error - Invalid syntax
api_error - External API failed
command_failed - Shell command failed
agent_failed - Subagent failed

Capability Categories

web_browsing - Internet access requests
image_processing - Image/photo handling
communication - Email/messaging
database_access - SQL/database queries
external_api - Third-party services
memory_persistence - Long-term memory

Configuration

In settings.json, these hooks enable self-improvement:

{
  "hooks": {
    "SessionStart": [...],  // Reviews learnings
    "PostToolUse": [
      {"matcher": "Bash", "hooks": [{"command": "enhanced-failure-logger.js"}]},
      {"matcher": "Task", "hooks": [{"command": "enhanced-failure-logger.js"}]}
    ]
  }
}

Benefits

Learn from mistakes - Don't repeat the same errors
Understand user preferences - Track what corrections mean
Identify skill gaps - Know what features to build
Improve over time - QAVR ranking gets better with feedback
Context persistence - Learnings survive session restarts

Integration with Other Skills

Skill	Integration
QAVR	Successful learnings boost memory Q-values
Memory Consolidation	Periodic cleanup of old learnings
Confidence Check	Review learnings before major tasks
IR-v2	Use learnings to inform pattern selection

Files

~/.claude/
├── logs/
│   ├── learnings.jsonl           # All learnings
│   ├── corrections.jsonl         # User corrections
│   ├── failures_detailed.jsonl   # Enhanced failure logs
│   ├── missing_capabilities.jsonl # Capability requests
│   └── capability_gaps.json      # Aggregated gaps
└── scripts/hooks/
    ├── correction-detector.js
    ├── enhanced-failure-logger.js
    ├── capability-tracker.js
    └── session-start.js (reviews learnings)

---name: self-improving-agentdescription: Enables continuous self-improvement through learning from failures, user corrections, and capability gaps. Integrates with QAVR for learned memory ranking.---# Self-Improving AgentCaptures learnings, errors, and corrections to enable continuous improvement across sessions.## Activation TriggersThis skill automatically activates when:| Trigger | What Gets Logged | Location ||---------|------------------|----------|| **Command fails** | Error type, context, recovery suggestion | `logs/failures_detailed.jsonl` || **User corrects** | "No, that's wrong...", "Actually..." | `logs/corrections.jsonl` || **Missing capability** | "Can you X?" where X isn't available | `logs/missing_capabilities.jsonl` || **API/tool fails** | Failure pattern, suggested fix | `logs/failures_detailed.jsonl` || **Better approach found** | Optimization learned | `logs/learnings.jsonl` |## How It Works### 1. Automatic Logging (via hooks)```PostToolUse → enhanced-failure-logger.js → logs failures with contextUserMessage → correction-detector.js → detects "wrong/actually/try again"Response → capability-tracker.js → detects unfulfilled requests```### 2. Learning AggregationAll learnings flow to `logs/learnings.jsonl`:```json{ "timestamp": "2026-01-26T12:00:00Z", "type": "user_correction|tool_failure|missing_capability", "category": "factual_error|command_failed|web_browsing", "description": "what was learned", "source": "which detector"}```### 3. Session Start ReviewOn each session start, recent learnings are shown:```=== Learning Review ===[Learnings] 47 total entries[Recent] • [tool_failure] Bash failed: timeout - WebFetch to external API... • [user_correction] User corrected: "No, use the other file..."[Corrections] 12 user corrections logged[Capability Gaps] Top requested: • send emails (5x) • browse web (3x)```### 4. QAVR IntegrationSuccessful learnings boost Q-values for related memories, improving future retrieval.## Manual Commands### Review Learnings```bash# Show all learningscat ~/.claude/logs/learnings.jsonl | tail -20# Show corrections onlycat ~/.claude/logs/corrections.jsonl | jq -s 'group_by(.correction_type) | map({type: .[0].correction_type, count: length})'# Show capability gaps reportnode ~/.claude/scripts/hooks/capability-tracker.js --report```### Test Detection```bash# Test correction detectornode ~/.claude/scripts/hooks/correction-detector.js# Test failure loggernode ~/.claude/scripts/hooks/enhanced-failure-logger.js# Test capability trackernode ~/.claude/scripts/hooks/capability-tracker.js```## Learning Categories### Correction Types- `factual_error` - Wrong information provided- `retry_request` - User asked to try again- `misunderstanding` - Misinterpreted the request- `failed_solution` - Solution didn't work### Failure Types- `permission_error` - Access denied- `not_found` - File/resource missing- `timeout` - Operation timed out- `network_error` - Connection issues- `syntax_error` - Invalid syntax- `api_error` - External API failed- `command_failed` - Shell command failed- `agent_failed` - Subagent failed### Capability Categories- `web_browsing` - Internet access requests- `image_processing` - Image/photo handling- `communication` - Email/messaging- `database_access` - SQL/database queries- `external_api` - Third-party services- `memory_persistence` - Long-term memory## ConfigurationIn `settings.json`, these hooks enable self-improvement:```json{ "hooks": { "SessionStart": [...], // Reviews learnings "PostToolUse": [ {"matcher": "Bash", "hooks": [{"command": "enhanced-failure-logger.js"}]}, {"matcher": "Task", "hooks": [{"command": "enhanced-failure-logger.js"}]} ] }}```## Benefits1. **Learn from mistakes** - Don't repeat the same errors2. **Understand user preferences** - Track what corrections mean3. **Identify skill gaps** - Know what features to build4. **Improve over time** - QAVR ranking gets better with feedback5. **Context persistence** - Learnings survive session restarts## Integration with Other Skills| Skill | Integration ||-------|-------------|| **QAVR** | Successful learnings boost memory Q-values || **Memory Consolidation** | Periodic cleanup of old learnings || **Confidence Check** | Review learnings before major tasks || **IR-v2** | Use learnings to inform pattern selection |## Files```~/.claude/├── logs/│ ├── learnings.jsonl # All learnings│ ├── corrections.jsonl # User corrections│ ├── failures_detailed.jsonl # Enhanced failure logs│ ├── missing_capabilities.jsonl # Capability requests│ └── capability_gaps.json # Aggregated gaps└── scripts/hooks/ ├── correction-detector.js ├── enhanced-failure-logger.js ├── capability-tracker.js └── session-start.js (reviews learnings)```

kimasplund/clawdbot-skills-pack/tree/main/self-improving-agent commit d60db35e98

Frequently asked questions

npx skillmds@latest add kimasplund/self-improving-agent