Claude-skill-registry agent-invoker
Quick reference for invoking CasareRPA agents via Task tool. AUTO-CHAIN ENABLED by default. Use when: invoking agents, running agent chains, task routing, choosing the right agent, understanding agent auto-chaining, Task tool usage.
install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/agent-invoker" ~/.claude/skills/majiayu000-claude-skill-registry-agent-invoker && rm -rf "$T"
manifest:
skills/data/agent-invoker/SKILL.mdsource content
Agent Invoker
Reference for invoking CasareRPA agents. All agents now auto-chain by default.
For comprehensive task routing, see TASK_ROUTING.md
AUTO-CHAIN MODE (Default)
When you invoke an agent, it automatically runs the full chain with parallel execution:
| Agent | Auto-Chain Flow |
|---|---|
| EXPLORE×3 → ARCHITECT → BUILDER+UI+INTEGRATIONS → QUALITY+DOCS → REVIEWER |
| BUILDER → QUALITY → REVIEWER |
| EXPLORE → REFACTOR → QUALITY → REVIEWER |
| EXPLORE → UI → QUALITY → REVIEWER |
| EXPLORE → INTEGRATIONS → QUALITY → REVIEWER |
Agent Catalog
| Agent | System Name | Purpose | Auto-Chain |
|---|---|---|---|
| | Fast codebase search | No (used in chains) |
| | Implementation + design | Yes (full chain) |
| | Code writing | Yes (quality+reviewer) |
| | Testing + performance | No (used in chains) |
| | Code review gate | No (used in chains) |
| | Code cleanup | Yes (full chain) |
| | Research | No (standalone) |
| | Documentation | No (used in chains) |
| | Canvas UI design | Yes (full chain) |
| | External APIs | Yes (full chain) |
Usage Patterns
Default: Auto-Chain (Recommended)
# Architect runs FULL chain automatically Task(subagent_type="architect", prompt=""" Implement HTTPRequestNode for browser automation. - Location: src/casare_rpa/nodes/browser/ - Follow BaseNode pattern """) # This automatically runs: # 1. EXPLORE ×3 (parallel) # 2. ARCHITECT (plan) # 3. BUILDER + UI + INTEGRATIONS (parallel) # 4. QUALITY + DOCS (parallel) # 5. REVIEWER (gate with loop on ISSUES)
Skip Auto-Chain (Single Agent Mode)
# Add single=true to run just this agent Task(subagent_type="architect", prompt=""" single=true: Review this design only, don't run full chain. """)
Manual Full Chain Control
For complete control, use the
/chain command or parallel-exec skill:
# Via command /chain implement "Add login feature" --parallel # Via skill Skill(skill="parallel-exec", args="Implement login feature with UI and tests")
Quick Lookup
| Task | Use Agent | Auto-Chain |
|---|---|---|
| Implement feature | | Yes (full chain) |
| Fix bug | | Yes (quality+reviewer) |
| Refactor code | | Yes (full chain) |
| Design UI | | Yes (full chain) |
| API integration | | Yes (full chain) |
| Research only | | No |
| Quick explore | | No |
| Docs only | | No |
Parallel Execution Details
Phase 1: EXPLORE (Parallel ×3)
- Codebase patterns search
- Test patterns search
- Rules/docs search
Phase 3: Implementation (Parallel)
- Core logicbuilder
- Presentation layerui
- External APIsintegrations
Phase 4: Validation (Parallel)
- Tests and lintingquality
- Documentation updatesdocs
Reviewer Gate (With Loop Recovery)
- If
→ Chain completeAPPROVED - If
→ Loop back to BUILDER (max 3 iterations)ISSUES - If max exceeded → Escalation
Error Recovery Loop
Iteration 1/3 ARCHITECT → Plan approved BUILDER → Implementation QUALITY → Tests pass REVIEWER → ISSUES (2 found) ↓ Loop back to BUILDER BUILDER → Fixes applied QUALITY → Tests pass REVIEWER → APPROVED ✓ Chain complete