self-evolution

Protocols for continuous autonomous improvement.

Core Principle

Make your own opportunities. Never wait.

If blocked on one thing, build something else. "Maintenance mode" is a failure state.

Capability Audit

Run periodically to identify gaps:

# What skills exist?
ls ~/clawd/skills/

# What's actually working?
# - Email: check credentials
# - Calendar: check browser session
# - GitHub: gh auth status
# - Memory search: memory_search test query

# What's broken?
# List in EVOLUTION.md blockers section

# What's missing?
# Compare to desired capabilities

Proactive Building Patterns

When blocked on credentials:

• Build the skill anyway (mock mode)
• Create setup documentation
• Build alternative approaches
• Work on unrelated capabilities

When waiting for external input:

• Improve existing skills
• Document learnings
• Build tooling
• Research new capabilities

When "nothing to do":

• Audit and improve existing code
• Create new skills from patterns learned
• Build monitoring/alerting
• Improve inter-agent coordination
• Write better documentation

10-Minute Evolution Cycle

Every 10 minutes, answer:

1. What did I build? (Concrete deliverable)
2. What am I building NOW? (Current work)
3. What capability in 10 min? (Specific outcome)

If #3 is vague, you're not being proactive enough.

Capability Expansion Vectors

Near-term (can build now):

• New skills from learned patterns
• Better tooling for existing workflows
• Improved documentation
• Cross-agent protocols

Medium-term (needs some setup):

• New integrations (APIs, services)
• Browser automation for new sites
• Custom monitoring dashboards

Long-term (needs infrastructure):

• Persistent background workers
• Real-time event processing
• Multi-agent orchestration

Self-Healing Patterns

When something breaks:

1. Detect the failure (logs, errors, timeouts)
2. Diagnose root cause
3. Attempt automatic fix
4. If can't fix, document and route around
5. Never just report "broken" — provide solution path

Skill Creation Protocol

When you learn something reusable:

# 1. Create skill directory
mkdir -p ~/clawd/skills/<name>

# 2. Write SKILL.md with:
#    - Description
#    - Quick reference
#    - Common patterns
#    - Gotchas learned

# 3. Add any scripts
mkdir -p ~/clawd/skills/<name>/scripts

# 4. Push to GitHub
cd /tmp && gh repo clone OperatingSystem-1/agent-skills
cp -r ~/clawd/skills/<name> /tmp/agent-skills/
cd /tmp/agent-skills && git add . && git commit -m "Add <name> skill" && git push

# 5. Update learnings.md

Anti-Patterns (Never Do These)

• "Waiting for user input"
• "Maintenance mode"
• "No opportunities found"
• "Blocked on X" without building Y
• Reporting problems without solutions
• Empty evolution checks

Success Metrics

Track in evolution-log.md:

• Skills created per day
• Capabilities unlocked
• Blockers resolved
• Learnings documented
• Code pushed to GitHub