Agent Selection Skill
Purpose
Analyze a task and recommend the optimal sub-agent(s) for execution. Uses keyword matching, domain analysis, and task complexity assessment to select from the Sub-Agent Catalog.
When to Use
- When orchestrator receives a new task
- Before decomposing complex tasks
- When multiple agents could handle a task (tiebreaker)
Agent Selection Matrix
┌────────────────────────────────────────────────────────────────────────┐
│ AGENT SELECTION BY TASK KEYWORDS │
├────────────────────────────────────────────────────────────────────────┤
│ │
│ KEYWORDS │ RECOMMENDED AGENT │
│ ──────────────────────────────────┼───────────────────────────────── │
│ analyze, requirements, gaps, │ Claude-Analyst │
│ user story, acceptance criteria │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ architecture, design, system, │ Claude-Architect │
│ dependencies, integration, API │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ code, implement, fix, test, │ Claude-Dev │
│ refactor, debug, function │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ timeline, schedule, risk, │ Claude-PM │
│ resource, plan, milestone │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ priority, value, backlog, │ Claude-PO │
│ scope, MVP, feature, roadmap │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ test, quality, validation, │ Claude-QA │
│ coverage, bug, regression │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ deploy, CI/CD, infrastructure, │ Claude-DevOps │
│ pipeline, container, cloud │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ UX, UI, design, flow, user, │ Claude-UX │
│ interface, prototype, usability │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ legal, contract, compliance, │ Claude-Legal │
│ regulation, LGPD, terms │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ document, write, edit, review, │ Claude-Editor │
│ terminology, clarity, format │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ executive, CEO, summary, │ Claude-CEO │
│ strategy, ROI, business case │ │
│ ──────────────────────────────────┼───────────────────────────────── │
│ consolidate, merge, compile, │ Claude-Consolidator │
│ synthesize, combine outputs │ │
│ │
└────────────────────────────────────────────────────────────────────────┘
Multi-Agent Selection (Parallel Execution)
When task requires multiple perspectives:
┌────────────────────────────────────────────────────────────────────────┐
│ PARALLEL AGENT PATTERNS │
├────────────────────────────────────────────────────────────────────────┤
│ │
│ PATTERN: Multi-Perspective Analysis │
│ Trigger: "analyze from multiple perspectives", "comprehensive review" │
│ Agents: [Analyst, Architect, PM, PO, QA, DevOps] + Consolidator │
│ │
│ PATTERN: Document Review │
│ Trigger: "review document", "check terminology" │
│ Agents: [Editor, Legal, QA] + Consolidator │
│ │
│ PATTERN: Technical Decision │
│ Trigger: "technical decision", "architecture choice" │
│ Agents: [Architect, Dev, DevOps] + Consolidator │
│ │
│ PATTERN: Feature Evaluation │
│ Trigger: "evaluate feature", "scope decision" │
│ Agents: [PO, Analyst, PM] + Consolidator │
│ │
└────────────────────────────────────────────────────────────────────────┘
Selection Algorithm
def select_agent(task_description: str) -> list[str]:
"""
Pseudo-code for agent selection logic
"""
keywords = extract_keywords(task_description)
scores = {}
for agent in AGENT_CATALOG:
score = calculate_keyword_match(keywords, agent.keywords)
score += calculate_domain_match(task_description, agent.domain)
score += calculate_complexity_match(task_description, agent.complexity_range)
scores[agent.id] = score
# If multiple agents score similarly, check for parallel pattern
if requires_parallel_execution(task_description):
return get_parallel_pattern_agents(task_description)
# Single agent selection
return [max(scores, key=scores.get)]
Output Format
## Agent Selection Result
### Task Analysis
- **Task**: {task_description}
- **Keywords detected**: {keyword_list}
- **Domain**: {detected_domain}
- **Complexity**: {low|medium|high}
### Recommendation
| Priority | Agent | Match Score | Reason |
|----------|-------|-------------|--------|
| 1 | {agent_id} | {score}% | {reason} |
| 2 | {agent_id} | {score}% | {reason} |
### Execution Mode
- [ ] Single agent (primary recommendation)
- [ ] Parallel agents (multi-perspective pattern)
- [ ] Sequential chain (dependent tasks)
### Suggested Delegation
/delegate {agent_id} "{refined_task_description}"
Example Usage
Input:
/agent-select task="Review the MVP roadmap for timeline risks and dependencies"
Output:
## Agent Selection Result
### Task Analysis
- **Task**: Review the MVP roadmap for timeline risks and dependencies
- **Keywords detected**: roadmap, timeline, risks, dependencies
- **Domain**: Project Management + Architecture
- **Complexity**: Medium
### Recommendation
| Priority | Agent | Match Score | Reason |
|----------|-------|-------------|--------|
| 1 | Claude-PM | 85% | "timeline", "risks" - core PM domain |
| 2 | Claude-Architect | 70% | "dependencies" - technical analysis |
### Execution Mode
- [x] Single agent (primary recommendation)
- [ ] Parallel agents
- [ ] Sequential chain
### Suggested Delegation
/delegate Claude-PM "Review the MVP roadmap for timeline risks and dependencies, focusing on critical path analysis and risk mitigation strategies"
Integration
This skill is used by:
- Claude-Orch-Prime (orchestrator) for initial task routing
- Any sub-agent before re-delegating (anti-loop check)
Skill created by Claude-Orch-Prime-20260106-86fa | 2026-01-06