Agent Skills: Delegate Skill

Delegate Skill

Automatically delegate tasks to the most appropriate specialized subagent.

Usage

/delegate "implement user authentication"
/delegate "review the payment module for security issues"
/delegate "write unit tests for the API endpoints"

How It Works

Step 1: Analyze Task Requirements

Parse the task description to identify:

• Domain: frontend, backend, API, database, security, testing, etc.
• Action: implement, review, test, refactor, analyze, document
• Technology: React, Django, Node, Python, etc. (if detectable from codebase)

Step 1.5: Load Coding Standards

Before delegating, auto-inject relevant coding standards based on the project's tech stack:

1. Read .claude/config/coding-standards.json for detection rules
2. Check TECHSTACK.md or codebase patterns to identify technologies
3. Load matching AGENTS.md files as additional context for the subagent

Detection Rules:

| Pattern Detected | Standards Loaded | |------------------|------------------| | WordPress, wp-content, Divi, Elementor | wordpress, php, mysql | | Next.js, App Router, use client | nextjs, react, javascript | | Laravel, Eloquent, Blade, Livewire | laravel, php, mysql | | Supabase, RLS, auth.uid() | supabase | | MariaDB, Galera | mariadb, mysql | | React, useState, JSX | react, javascript | | PHP, composer.json | php | | MySQL, my.cnf | mysql |

Include in Subagent Prompt:

## Coding Standards

Follow these coding standards for this task. Focus on CRITICAL and HIGH impact rules:

[Include relevant AGENTS.md content here]

Key rules to follow:
- [List 3-5 most relevant CRITICAL rules from loaded standards]

Example: If TECHSTACK.md mentions "Next.js" and "Supabase":

• Load: setup/skills/nextjs-best-practices/AGENTS.md
• Load: setup/skills/react-best-practices/AGENTS.md
• Load: setup/skills/javascript-best-practices/AGENTS.md
• Load: setup/skills/supabase-best-practices/AGENTS.md

Step 2: Check Parallel Execution Slots

Read CLAUDE-activeContext.md to check parallel execution tracking:

## Parallel Execution Tracking

**Maximum Concurrent Agents:** 2

| Slot | Agent | Task | Status |
|------|-------|------|--------|
| 1 | [agent] | [task] | Active/Available |
| 2 | [agent] | [task] | Active/Available |

Rules:

• If both slots are Active, QUEUE the task and notify user
• If at least one slot is Available, proceed with delegation

Step 3: Select Appropriate Agent

Agent Selection Hierarchy

1. Exact match: Task mentions specific technology → use specialist
2. Domain match: Task domain matches agent specialty → use domain expert
3. Universal fallback: No specialist → use general-purpose agent

Available Agents (from setup/agents/critical/)

| Agent | Specialty | Use When | |-------|-----------|----------| | software-engineer | Implementation | Building features, writing code | | code-reviewer | Quality | Reviewing code for issues | | qa-engineer | Testing | Writing tests, quality assurance | | product-manager | Requirements | Clarifying scope, priorities |

Phase-Based Agent Assignment

If current phase is known (from CLAUDE-activeContext.md), prefer phase-appropriate agent:

| Phase | Preferred Agent | |-------|-----------------| | 2-3 | Requirements Analyst / Solution Architect | | 4-5 | Technical Planner / Software Engineer | | 6 | QA Engineer | | 7 | Security Auditor | | 8 | Code Reviewer | | 9 | DevOps Engineer |

Step 4: Launch Subagent

Use the Task tool to launch the selected agent:

Task:
  description: "[3-5 word summary]"
  prompt: |
    ## Context
    [Current phase and project context from CLAUDE-activeContext.md]

    ## Task
    [Full task description from user]

    ## Coding Standards
    Follow these coding standards for this task. Focus on CRITICAL and HIGH impact rules:

    [Include relevant AGENTS.md content based on detected tech stack]

    Key rules to enforce:
    - [List 3-5 most relevant CRITICAL rules]

    ## Constraints
    - Follow patterns in CLAUDE-patterns.md
    - Follow coding standards loaded above
    - Update CLAUDE-activeContext.md when complete
    - Report findings back to main agent

    ## Deliverables
    [Expected outputs]
  subagent_type: [agent-type]

Step 5: Update Tracking

Update CLAUDE-activeContext.md:

## Active Tasks

| ID | Task | Agent | Status | Started |
|----|------|-------|--------|---------|
| T-001 | [task summary] | [agent] | In Progress | [timestamp] |

## Parallel Execution Tracking

| Slot | Agent | Task | Status |
|------|-------|------|--------|
| 1 | [agent] | [task] | Active |
| 2 | - | - | Available |

Agent Matching Examples

Example 1: Security Task

Input: "review the API for SQL injection vulnerabilities"

Analysis:
- Domain: security
- Action: review
- Technology: API/backend

Selection: Security Auditor (if available) or Code Reviewer
Reasoning: Security-focused review requires security expertise

Example 2: Implementation Task

Input: "implement user registration with email verification"

Analysis:
- Domain: backend
- Action: implement
- Technology: authentication

Selection: Software Engineer
Reasoning: Implementation task for new feature

Example 3: Testing Task

Input: "write integration tests for the checkout flow"

Analysis:
- Domain: testing
- Action: write tests
- Technology: e-commerce

Selection: QA Engineer
Reasoning: Testing task requires QA expertise

Example 4: Parallel Limit Reached

Current State:
- Slot 1: Active (software-engineer working on auth)
- Slot 2: Active (qa-engineer writing tests)

Input: "refactor the database models"

Response:
"Cannot delegate now - maximum 2 agents already running.
Current tasks:
1. software-engineer: implementing auth
2. qa-engineer: writing tests

Task queued. Will delegate when a slot becomes available."

Subagent Types Mapping

Map agents to Claude Code subagent types:

| Agent | subagent_type | Model | |-------|---------------|-------| | software-engineer | Explore or general-purpose | sonnet | | code-reviewer | Explore | sonnet | | qa-engineer | general-purpose | sonnet | | code-searcher | Explore | sonnet | | product-manager | Plan | sonnet |

Task Completion Handling

When a subagent completes:

1. Update CLAUDE-activeContext.md:

   • Change task status to "Completed"
   • Free the parallel execution slot
   • Add completion entry to Session Log

2. Check queued tasks:

   • If tasks are queued, delegate the next one

3. Report results:

   • Summarize findings to user
   • Suggest next actions if applicable

Error Handling

• No matching agent: Use general-purpose subagent with detailed context
• Agent definition not found: Fall back to built-in subagent types
• Memory bank missing: Create CLAUDE-activeContext.md with default structure