agent-invoker

Quick reference for invoking CasareRPA agents via the Task tool.

IMPORTANT: Agent Name Mapping

The Task tool uses system agent names, not custom names. Always use the system name:

Agent Catalog

Invocation Patterns

Exploration (Read-Only)

# Quick file search (uses system name: Explore)
Task(subagent_type="Explore", prompt="""
Find all files matching pattern: src/**/*node*.py
Focus: Browser automation nodes
""")

# Code search
Task(subagent_type="Explore", prompt="""
Search for: "async def execute"
Scope: src/casare_rpa/nodes/
Return: File paths and line numbers
""")

Implementation Flow

# Step 1: Architect implements (system: rpa-engine-architect)
Task(subagent_type="rpa-engine-architect", prompt="""
Implement HTTPRequestNode for browser automation.
- Location: src/casare_rpa/nodes/browser/
- Follow BaseNode pattern
- Include proper error handling
""")

# Step 2: Quality tests (system: chaos-qa-engineer)
Task(subagent_type="chaos-qa-engineer", prompt="""
mode: test
Create test suite for HTTPRequestNode.
Cover: success, errors, edge cases.
Location: tests/nodes/browser/test_http_node.py
""")

# Step 3: MANDATORY Review (system: code-security-auditor)
Task(subagent_type="code-security-auditor", prompt="""
Review HTTPRequestNode implementation:
- src/casare_rpa/nodes/browser/http_node.py
- tests/nodes/browser/test_http_node.py

Output: APPROVED or ISSUES (with file:line references)
""")

Review Loop Pattern

# If reviewer returns ISSUES, fix and re-review:

# architect fixes
Task(subagent_type="rpa-engine-architect", prompt="""
Fix reviewer issues:
1. src/http_node.py:45 - Add timeout handling
2. src/http_node.py:78 - Fix type hint

Re-submit for review after fixing.
""")

# quality re-runs tests
Task(subagent_type="chaos-qa-engineer", prompt="""
mode: test
Re-run tests after fixes.
Verify all pass.
""")

# reviewer re-reviews
Task(subagent_type="code-security-auditor", prompt="""
Re-review after fixes:
- src/casare_rpa/nodes/browser/http_node.py
- tests/nodes/browser/test_http_node.py
""")

Parallel Execution

# Launch multiple independent tasks in parallel
Task(subagent_type="Explore", prompt="Find browser node patterns...")
Task(subagent_type="Explore", prompt="Find test fixtures...")
Task(subagent_type="Explore", prompt="Find similar implementations...")

# Or mix different agents (use system names!)
Task(subagent_type="rpa-engine-architect", prompt="Implement feature A...")
Task(subagent_type="rpa-docs-writer", prompt="Document feature B...")
Task(subagent_type="chaos-qa-engineer", prompt="Test feature C...")

Specialized Agents

# Security audit (system: security-architect)
Task(subagent_type="security-architect", prompt="""
Audit credential handling in Robot executor.
Focus: secrets, sandboxing, input validation.
""")

# Research (system: rpa-research-specialist)
Task(subagent_type="rpa-research-specialist", prompt="""
Compare UiPath vs CasareRPA selector strategies.
Recommend approach for desktop automation.
""")

# UI Design (system: rpa-ui-designer)
Task(subagent_type="rpa-ui-designer", prompt="""
Design properties panel for IfNode.
Include: condition editor, true/false labels.
""")

# Performance (system: chaos-qa-engineer with mode)
Task(subagent_type="chaos-qa-engineer", prompt="""
mode: perf
Profile workflow execution for 100-node workflows.
Identify bottlenecks.
""")

# Refactoring (system: rpa-refactoring-engineer)
Task(subagent_type="rpa-refactoring-engineer", prompt="""
Break up MainWindow (1200 lines) into focused controllers.
Follow controller pattern from systemPatterns.md.
""")

Agent Modes

Quality Agent Modes (chaos-qa-engineer)

# Testing mode (default)
Task(subagent_type="chaos-qa-engineer", prompt="mode: test\n...")

# Performance mode
Task(subagent_type="chaos-qa-engineer", prompt="mode: perf\n...")

# Stress testing mode
Task(subagent_type="chaos-qa-engineer", prompt="mode: stress\n...")

Common Mistakes

Wrong

# Don't skip the reviewer
Task(subagent_type="architect", prompt="Implement and merge...")  # NO!

# Don't use wrong agent for task
Task(subagent_type="docs", prompt="Fix this bug...")  # Use architect!

# Don't forget .brain protocol
Task(subagent_type="architect", prompt="Just implement...")  # Missing context!

Right

# Always follow the flow: architect -> quality -> reviewer
# Always reference .brain/ for context
# Use explore for research, architect for implementation

Quick Lookup