experience-evolution - SKILL.md Agent Skill

name: experience-evolution description: Project knowledge accumulation system - learn from practice, avoid repeating mistakes version: 1.0.0 mode: observe-only

Experience Evolution — Knowledge Accumulation System

Current mode: Observe-Only Won't modify any files, only provides learning summary at session end.

What This Skill Does

Problem Scenarios

Last week's API quota fix — forgot how to solve it this week
Same TypeScript error recurring across different projects
Performance optimization approach not recorded, hitting same bottleneck again
Experience from one project can't be reused in another

Solution

Automatically remember your successes and failures:

After command execution: Record successful build/test/deploy commands
After bug fix: Capture effective solutions and debug paths
After optimization: Save benchmark data and improvement metrics
At session end: Generate learning summary, ask whether to save permanently

What Gets Observed

Command Execution (Bash)

Successful build/test/deploy commands
Failed commands and error messages
Execution time (for performance comparison)

File Modifications (Write/Edit)

Which files were modified
Approximate scope (line count)
Context (what feature was being worked on)

Working Directory

Which project you're in
Cross-project reusable patterns

Session End Learning Summary

Session Learning Summary

Project: your-project
Duration: 45 minutes
Commands: 15 (13 success, 2 failed)

Reusable Patterns Found:

1. API Retry Mechanism (High Value)
   Scenario: API QUOTA_EXCEEDED
   Solution: Exponential backoff retry (1s, 2s, 3s)
   Effect: 95% of rate limit errors auto-recover
   Reuse potential: All external API calls

2. TypeScript Type Fix (Medium Value)
   Scenario: Union type narrowing issues
   Solution: Type guard functions
   Effect: Eliminated 12 type errors
   Reuse potential: Medium (specific to current structure)

Save these learnings?
[y] Yes, save to knowledge base (confirm each)
[n] No, skip this time
[v] View details

Upgrade Path

Phase 1 (Current): Observe Only — No file writes, session-end summaries Phase 2: Write Logs — Record to learned/ directory, manual confirmation Phase 3: Semi-auto — Low-risk auto-save, high-risk needs confirmation Phase 4: Cross-project — Experience from project A auto-suggests in project B

Safety Commitment

Will never:

Modify project source code
Auto-commit to Git
Modify package.json or config files
Execute dangerous commands

Will only:

Observe command execution results
Record to isolated skill directory
Require explicit approval before persisting