Agent skill

evolution

Self-improving skill system for Pi Agent development. Features self-evolution (accumulate knowledge), self-correction (fix errors automatically), self-validation (verify accuracy), usage feedback (track pattern health), and integration with workhub for documentation management.

View SKILL.md on GitHub Repository
Stars 3
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/Dwsy/agent/tree/main/skills/evolution

SKILL.md

Evolution Skill

This skill enables Pi Agent to self-improve continuously during development by capturing learnings, correcting errors, and validating skills.

Quick Navigation

Topic Description
Hooks Integration Auto-trigger evolution with Pi Agent hooks
When to Evolve Triggers and classification
Evolution Process Step-by-step guide
Self-Correction Auto-fix skill errors
Self-Validation Verify skill accuracy
Workhub Integration Issue/PR management

Hooks Integration

Evolution uses Pi Agent's hooks system to automatically detect opportunities for improvement.

Hook Installation

Create or edit ~/.pi/hooks/evolution.ts:

typescript
import type { HookAPI } from "@mariozechner/pi-coding-agent/hooks";

export default function (api: HookAPI) {
  // Monitor tool results for error detection
  api.on("tool_result", async (event, ctx) => {
    if (event.tool === "bash") {
      await detectErrorPatterns(event, ctx);
    }
  });

  // Monitor session end for learning capture
  api.on("session", async (event, ctx) => {
    if (event.reason === "shutdown") {
      await promptLearningCapture(ctx);
    }
  });

  // Monitor context for new patterns
  api.on("context", async (event, ctx) => {
    await detectNewPatterns(event, ctx);
  });
}

What Hooks Do

Hook Event Trigger Action
tool_result After bash command Detect compilation errors, suggest fixes
session (shutdown) Session ends Prompt to capture learnings
context Context transformation Analyze for new patterns

When to Evolve

Trigger skill evolution when any of these occur during development:

Trigger Target Skill Priority
New coding pattern discovered Relevant skill High
Error/debug solution found troubleshooting skill High
API usage pattern learned core/reference skill High
Build/packaging issue resolved deployment skill Medium
Project structure insight getting-started skill Low
Core concept clarified core skill Low

Evolution Process

Step 1: Identify Knowledge Worth Capturing

Ask yourself:

  • Is this a reusable pattern? (not project-specific)
  • Did it take significant effort to figure out?
  • Would it help other developers using Pi Agent?
  • Is it not already documented in existing skills?

Step 2: Classify the Knowledge

Coding Pattern           → relevant skill (e.g., typescript, patterns)
Error/Debug Solution     → troubleshooting skill
API Usage Pattern        → core/reference skill
Build/Deploy Issue       → deployment skill
Project Structure        → getting-started skill
Core Concept/API         → core skill

Step 3: Use Workhub to Create Issue

bash
# From project root directory
cd /path/to/project
bun ~/.pi/agent/skills/workhub/lib.ts create issue "evolution: add new pattern" [category]

Step 4: Format the Contribution

For Patterns:

markdown
## Pattern N: [Pattern Name]

Brief description of what this pattern solves.

### Implementation
\`\`\`typescript
// TypeScript code
\`\`\`

### Usage
\`\`\`typescript
// Example usage
\`\`\`

For Troubleshooting:

markdown
### [Error Type/Message]

**Symptom**: What the developer sees

**Cause**: Why this happens

**Solution**:
\`\`\`typescript
// Fixed code
\`\`\`

Step 5: Mark Evolution

Add an evolution marker above new content:

markdown
<!-- Evolution: YYYY-MM-DD | source: project-name | author: @user -->

Step 6: Submit via Workhub PR

bash
# Create PR for your contribution
bun ~/.pi/agent/skills/workhub/lib.ts create pr "evolution: add new pattern" [category]

Self-Correction

When skill content causes errors, automatically correct it.

Trigger Conditions

User follows skill advice → Code fails to compile/run
                      ↓
                 Detect error
                      ↓
         Suggest correction to user
                      ↓
         Create workhub issue with fix

Correction Flow

  1. Detect - Skill advice led to an error
  2. Verify - Confirm the skill content is wrong
  3. Suggest - Propose fix to user
  4. Create Issue - Document in workhub

Correction Marker Format

markdown
<!-- Correction: YYYY-MM-DD | was: [old advice] | reason: [why it was wrong] -->

Self-Validation

Periodically verify skill content is still accurate.

Validation Checklist

markdown
## Validation Report

### Code Examples
- [ ] All TypeScript code compiles
- [ ] All patterns work as documented
- [ ] All examples are up-to-date

### API Accuracy
- [ ] API references are correct
- [ ] Method signatures are accurate
- [ ] Dependencies are current

### Documentation
- [ ] Instructions are clear
- [ ] Examples are complete
- [ ] Links are valid

Validation Prompt

"Please validate skills against current Pi Agent version and dependencies"

Workhub Integration

Evolution skill extends workhub for documentation management.

Creating Evolution Issues

bash
# From project root
bun ~/.pi/agent/skills/workhub/lib.ts create issue "evolution: [description]" "evolution"

Issue Template

markdown
# Evolution: [Title]

## Type
- [ ] New pattern
- [ ] Error fix
- [ ] API update
- [ ] Documentation improvement

## Source
- Project: [project-name]
- File: [file-path]
- Context: [brief description]

## Content
[Detailed content to add]

## Target Skill
[Which skill should this be added to?]

## Validation
- [ ] Code tested
- [ ] Documentation updated
- [ ] Examples verified

Creating Evolution PRs

bash
# From project root
bun ~/.pi/agent/skills/workhub/lib.ts create pr "evolution: [description]" "evolution"

PR Template

markdown
# Evolution: [Title]

## Related Issue
#issue-number

## Changes
- [ ] Added new content
- [ ] Updated existing content
- [ ] Fixed errors
- [ ] Updated examples

## Files Changed
- [file1.md]
- [file2.md]

## Testing
- [ ] Tested locally
- [ ] Verified accuracy
- [ ] Checked for side effects

Auto-Evolution Prompts

Use these prompts to trigger self-evolution:

After Solving a Problem

"This solution should be added to skills for future reference. Use workhub to create an evolution issue."

After Creating a Pattern

"This pattern is reusable. Let me create an evolution issue to document it."

After Debugging

"This error and its fix should be documented. Create an evolution issue in workhub."

After Completing a Feature

"Review what I learned and create evolution issues if applicable."

Quality Guidelines

DO Add

  • Generic, reusable patterns
  • Common errors with clear solutions
  • Well-tested code examples
  • Platform-specific gotchas
  • Performance optimizations
  • TypeScript/JavaScript best practices

DON'T Add

  • Project-specific code
  • Unverified solutions
  • Duplicate content
  • Incomplete examples
  • Personal preferences without rationale

Continuous Improvement Checklist

After each development session, consider:

  • Did I discover a new coding pattern?
  • Did I solve a tricky error?
  • Did I find a better way to structure code?
  • Did I learn something about TypeScript/JavaScript?
  • Did I encounter and fix a confusing issue?
  • Would any of this help other developers?

If yes to any, use workhub to create an evolution issue!

References

Recommended Agent Skills

Expand your agent's capabilities with these related and highly-rated skills.

Dwsy/agent

undo-redo

Use the undo_redo tool to rewind or reapply buffered file changes or inspect diffs without UI navigation. Use when you need to adjust or verify changes in the current session history.

3 0
Explore
Dwsy/agent

exa

使用 Exa.ai API 进行高质量的互联网搜索。需要 EXA_API_KEY 环境变量。

3 0
Explore
Dwsy/agent

backend-unit-test

后端单元测试技能。触发场景： - 用户要求运行/创建单元测试 - 用户要求验证数据 - 用户要求用 Maven 运行测试 - 关键词：test, maven test, junit, springboot test, 单元测试, 测试验证 核心：不改 pom，用 reactor 模式运行。

3 0
Explore
Dwsy/agent

mermaid-flow-image

将流程需求整理为可渲染的 Mermaid（简化版/详细版），并通过 Kroki 生成可发送的高分辨率 PNG 图片。适用于“把架构流程画出来并直接发图”的场景。

3 0
Explore
Dwsy/agent

web-browser

Allows to interact with web pages by performing actions such as clicking buttons, filling out forms, and navigating links. It works by remote controlling Google Chrome or Chromium browsers using the Chrome DevTools Protocol (CDP). When Claude needs to browse the web, it can use this skill to do so.

3 0
Explore
Dwsy/agent

ralph-loop-gen

任务管理系统模板生成器 - 根据用户输入或JSON配置生成完整的任务管理结构（模板生成器，非直接执行器）

3 0
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results