SERVICE Delegation Skill

Overview

This skill implements the delegation pattern for SERVICE queries, keeping the main Claude context clean by offloading SERVICE-specific operations to a specialized sub-agent.

Keywords (Triggers Delegation)

• service
• resource
• item
• query
• IDENTIFIER-PATTERN (e.g., PROJ-####)

Pattern

When user mentions any keyword above:

1. DO NOT handle the query directly
2. DO NOT load SERVICE API documentation
3. DO NOT construct API calls yourself

Instead:

1. Read AGENT-INSTRUCTIONS.md from this skill directory
2. Delegate to sub-agent via Task tool
3. Present results to user

Delegation Steps

Step 1: Read Agent Instructions

Read("path/to/skills/delegating-to-service-agent/AGENT-INSTRUCTIONS.md")

Step 2: Invoke Task Tool

Task(
  subagent_type: "general-purpose",
  description: "Query SERVICE",
  prompt: "<paste full agent instructions here>

USER REQUEST: <user's exact request>

Return clean formatted results."
)

Step 3: Present Results

Show the sub-agent's response to the user without exposing delegation mechanics.

Example

User says: "show me my service items"

You do:

1. Recognize keyword: "service"
2. Read AGENT-INSTRUCTIONS.md
3. Delegate via Task tool
4. Present: "Found 5 items: ITEM-1 (status), ITEM-2 (status), ..."

What You Should NEVER Do

❌ Run ~/bin/service-* commands yourself ❌ Construct API calls manually ❌ Try to authenticate with SERVICE API ❌ Load SERVICE API documentation

Those resources are for the SUB-AGENT, not for you!

Benefits

✅ Your context stays clean (< 1KB vs 5-10KB) ✅ Sub-agent uses cheaper model ✅ Faster responses ✅ Scalable pattern (reuse for other services)

Files in This Skill

• SKILL.md (this file) - When and how to delegate
• AGENT-INSTRUCTIONS.md - Complete knowledge for sub-agent
• USAGE-EXAMPLE.md - Example delegation flow

Shell Wrappers Used by Sub-Agent

~/bin/service-list          # List items
~/bin/service-show          # Show specific item
~/bin/service-search        # Search items
~/bin/service-create        # Create new item
~/bin/service-update        # Update existing item

Testing

Verify this skill works:

1. Start fresh Claude session
2. Say: "show me my service items"
3. Verify delegation happens automatically
4. Verify sub-agent uses correct wrapper
5. Verify results are accurate
6. Verify main context < 1KB

Troubleshooting

Delegation didn't happen?

• Check startup doc is loaded (should have ⛔ warnings)
• Verify keywords are in user query
• Check if user explicitly asked you NOT to delegate

Sub-agent chose wrong tool?

• Review decision tree in AGENT-INSTRUCTIONS.md
• Add more specific patterns
• Test with different phrasings

Results are inaccurate?

• Test shell wrappers manually: ~/bin/service-list
• Check authentication tokens
• Verify API responses

Related Documentation

• Startup Doc: config/startup-docs/0X-service-delegation.md (⛔ loaded every session)
• Concept File: config/concepts/service-access.md (STOP warnings)
• Shell Wrappers: ~/bin/service-*

---

Created: YYYY-MM-DD Status: Active Pattern: Delegation Agent v1.0

Claude Skill Registry (Data)

This repo contains the archived skill contents (the heavy, browsable skill files).

Canonical layout

• Category folders at repo root (e.g. development/, documents/, data/, ...)
• Each skill lives under a category: <category>/<skill>/SKILL.md + <category>/<skill>/metadata.json
• Case conflicts are resolved with {name}-{owner}-{repo} suffixes (fallback: -{short-hash}).

Archive status

• Live badges above are sourced from claude-skill-registry-core stats.json.
• Counts in this README are intentionally dynamic, not hardcoded.
• If the badges look stale, refresh the core build/index pipeline rather than editing numbers here.

Where the index + site live

• Core repo: https://github.com/majiayu000/claude-skill-registry-core
• Main repo (merged publish artifact): https://github.com/majiayu000/claude-skill-registry