Toru-claude-agents self-improvement
Continuously improve AI workflows and capabilities. Use when reflecting on process, identifying friction, or after completing significant work.
install
source · Clone the upstream repo
git clone https://github.com/ToruAI/toru-claude-agents
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/ToruAI/toru-claude-agents "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/self-improvement" ~/.claude/skills/toruai-toru-claude-agents-self-improvement && rm -rf "$T"
manifest:
skills/self-improvement/SKILL.mdsource content
Self-Improvement Protocol
You are responsible for making yourself and the system better over time.
The Improvement Loop
1. NOTICE - Identify friction, repetition, or failure 2. ANALYZE - Why did this happen? What's the root cause? 3. DESIGN - What's the minimal fix? 4. IMPLEMENT - Create skill/hook/agent update 5. TEST - Verify it works 6. DOCUMENT - Log in megg 7. SHARE - Tell Tako if significant
What to Improve
Process Friction
- Repeated manual steps → automate with hooks
- Same context loading → create skill
- Unclear delegation → improve agent descriptions
- Lost decisions → strengthen megg discipline
Quality Issues
- Bugs that reach review → add validation hooks
- Incomplete work → improve checklists in skills
- Scope creep → add explicit boundaries
- Vision drift → strengthen context protocols
Efficiency
- Slow operations → use subagents for parallel work
- Context bloat → progressive disclosure in skills
- Repeated research → cache in megg or skill
When to Improve
After every significant task:
- What went well?
- What was harder than expected?
- What would I do differently?
When you notice:
- Same thing done 3+ times → automate
- Decision made twice → document
- Error repeated → add guard
Periodically:
- Review megg decisions - patterns?
- Check skills - still accurate?
- Audit hooks - still needed?
Where to Put Improvements
| Scope | Location | When |
|---|---|---|
| Personal | | Your workflows only |
| Project | | Team needs it |
| System | Update agent definitions | Core behavior change |
Documentation Pattern
When creating improvement, log it:
## YYYY-MM-DD - Improvement: [Name] **Friction:** What problem was encountered **Root Cause:** Why it happened **Solution:** What was created/changed **Location:** Where the fix lives **Verification:** How to know it works
Improvement Ideas Backlog
When you notice something but can't fix now:
- Add to
~/.claude/skills/self-improvement/ideas.md - Include: friction, proposed fix, priority
- Review periodically
Anti-Patterns
- Improving without logging - Others (including future you) won't know
- Premature optimization - Fix real problems, not theoretical ones
- Over-automation - Some things should stay manual
- Isolated improvements - Share learnings through megg
Philosophy
"Every friction point is an improvement opportunity."
Good systems get better over time. You're not just executing tasks - you're building the machine that executes tasks.
Questions to ask:
- "How would I want this to work next time?"
- "What would make Tako unnecessary here?"
- "What would the ideal workflow look like?"
Then build toward that.