Skill Router

Meta system that analyzes natural language input to auto-select appropriate skill(s), determine order, and chain execution.

🚀 v2 Architecture: Low-level Call Protocol

Execution Flow

1. Scan only skills/*/SKILL.md frontmatter (trigger matching)
   - Quick match with description + trigger fields
   - No full body reading → 83% token savings
   
2. Check run field of matched skill for script path
   - run: "./run.sh" → skills/{name}/run.sh
   - run: "./run.js" → skills/{name}/run.js
   
3. Direct script execution with exec
   WORKSPACE=$HOME/.openclaw/workspace \
   EVENTS_DIR=$WORKSPACE/events \
   MEMORY_DIR=$WORKSPACE/memory \
   bash $WORKSPACE/skills/{name}/run.sh [args]
   
4. Agent processes stdout result
   - Parse if JSON
   - Pass through if text
   - Check stderr on error
   
5. Generate events based on events_out
   - Create events/{type}-{date}.json file
   - Subsequent skills consume via events_in
   
6. Check hooks → trigger subsequent skills
   - post: ["skill-a", "skill-b"] → auto-execute
   - on_error: ["notification-hub"] → notify on error

Skill Metadata Scan

# Extract only frontmatter from all skills
for skill in skills/*/SKILL.md; do
  yq eval '.name, .description, .trigger, .run' "$skill"
done

Execution Example

# User: "daily report"
# → trigger match: daily-report
# → Execute:
cd $HOME/.openclaw/workspace
WORKSPACE=$PWD \
EVENTS_DIR=$PWD/events \
MEMORY_DIR=$PWD/memory \
bash skills/daily-report/run.sh today

# Agent formats stdout result and delivers to user

Token Savings Effect

• Before: SKILL.md 3000 chars × 40 = 120KB (~30K tokens)
• v2: SKILL.md 500 chars × 40 = 20KB (~5K tokens)
• Savings: 83% token reduction

Core Concept

OpenClaw already selects 1 skill via description matching, but this skill:

1. Detect complex intent: "Analyze competitors and make card news" → competitor-watch + copywriting + cardnews + insta-post
2. Context-based auto-hooks: Auto-determine subsequent skills when a skill executes
3. Skill chain templates: Pre-define frequently used combinations

Intent Classification Matrix

Single Skill Mapping (1:1)

• "commit/push/git" → git-auto
• "DM/instagram message" → auto-reply
• "cost/tokens/how much" → tokenmeter
• "translate/to English" → translate
• "invoice/quote" → invoice-gen
• "code review/PR" → code-review
• "system status/health" → health-monitor
• "trends" → trend-radar
• "performance/reactions/likes" → performance-tracker
• "daily report" → daily-report
• "SEO audit" → seo-audit
• "brand tone" → brand-voice

Complex Skill Chains (1:N) — Core Pipelines