# Improve AI Skills With Learning Memory

AI agents often repeat the same mistakes across tasks. This skill records lessons, extracts patterns, and guides future skill updates.

## Install

```bash
npx skillstore add charon-fan/self-improving-agent
```

## Metadata

- - Slug: charon-fan-self-improving-agent
- - Version: 1.0.0
- - Author: charon-fan
- - GitHub username: charon-fan
- - License: MIT
- - Repository: https://github.com/charon-fan/agent-playbook/tree/main/skills/self-improving-agent/
- - Ref: main
- - Supported tools: Claude, Codex, Claude Code
- - Risk level: high
- - Risk factors: external\_commands, network, filesystem
- - Quality score: 38
- - Quality tier: warning
- - Public page: https://skillstore.pages.dev/skills/charon-fan-self-improving-agent
- - Manifest: https://skillstore.pages.dev/api/skills/charon-fan-self-improving-agent/manifest

## Capabilities

- Defines semantic, episodic, and working memory structures for agent learning.
- Guides extraction of reusable patterns from completed skill sessions.
- Provides workflows for correcting inaccurate skill guidance after errors.
- Documents evolution markers that trace skill changes to source episodes.
- Includes optional Claude Code hook examples for session and Bash event logging.

## Use Cases

- Improve Team Agent Playbooks: Convert repeated task lessons into reusable skill guidance with source markers.
- Track Debugging Lessons: Record recurring fixes and promote reliable patterns into debugger guidance.
- Review Skill Quality Over Time: Use validation templates and confidence scores to decide which patterns remain useful.

## Prompt Templates

### Summarize One Session

```
Review the latest skill session. List what happened, what worked, what failed, and one reusable lesson.
```

### Extract Reusable Patterns

```
Compare these recent sessions. Extract patterns that apply across tasks, then assign confidence and target skills.
```

### Correct Skill Guidance

```
Find the guidance that contributed to this failure. Propose a correction marker, validation steps, and affected skills.
```

### Audit Learning Memory

```
Audit semantic and episodic memory. Identify stale patterns, conflicting guidance, missing evidence, and updates needing human approval.
```

## Limitations

- Requires broad file-editing permissions to update skills and memory files.
- Hook examples can expose sensitive tool input or command output in logs.
- Does not include an independent validation engine for learned patterns.
- Research links and examples may become outdated over time.

## Best Practices

- Review every proposed skill change before applying it to shared repositories.
- Keep confidence scores tied to clear evidence and user feedback.
- Disable hook logging when tool input or output may contain secrets.

## Anti Patterns

- Do not auto-apply broad skill changes from a single weak example.
- Do not log command output from sensitive environments.
- Do not treat research citations as proof that a pattern is valid.

## Security Audit

- - Safe to publish: false
- - Audited at: 2026-06-28T21:47:21.769\+00:00
- - Summary: Static analysis produced many alerts, but most blocker-level items are false positives from markdown examples, diagrams, and research links. The confirmed risk is high because this community skill requests broad file-editing and Bash capabilities, teaches self-modification of skills, and documents hooks that can log tool input and command output.

## Stats

- - Views: 232
- - Downloads: 118
- - Favorites: 3
- - Popularity score: 0