coordinate

/coordinate

Claude Code Native Adaptation

Use Task tool + TaskCreate for task tracking and agent spawning instead of CLI spawning.

Response language follows language setting in .agents/config/user-preferences.yaml.

Step 1: Requirements Analysis

• Analyze user requirements
• Identify relevant domains (based on skill-routing.md)
• Use code analysis (Grep/Glob/Read) to understand codebase structure around affected areas

Step 2: PM Task Decomposition

• Complex requirements: Spawn pm-planner subagent via Task tool
• Simple requirements: Decompose tasks inline
• Save plan to .agents/plan.json

Step 3: User Confirmation

• Present plan to user
• MUST get user confirmation before proceeding

Step 4: Create Tasks with TaskCreate + Spawn Agents

Create each agent task with TaskCreate for progress tracking:

1. Register each task with TaskCreate (subject, description)
2. Spawn agents by priority tier (parallel)
   - Same tier agents: Multiple Task tool calls in same message
3. Update status with TaskUpdate on completion

Agent mapping: Reference .claude/agents/{agent}.md

Step 5: Monitoring

Task tool returns results directly → No polling needed.

Step 6: QA Review

Spawn qa-reviewer subagent via Task tool (review results: OWASP Top 10, performance, accessibility, code quality)

Step 7: Issue Remediation Loop (Native Loop)

LOOP:
  [1] Spawn implementation agents via Task tool (parallel)
  [2] Receive results
  [3] Spawn QA agent via Task tool (review results)
  [4] Parse QA results:
      - 0 CRITICAL/HIGH issues → ACCEPT ✓
      - CRITICAL/HIGH issues exist:
        → Compose issues as feedback for that agent:
          ## Issue Remediation
          **Agent**: {agent-name}
          **Issues**:
          - [CRITICAL] file:line — description — fix instruction
          - [HIGH] file:line — description — fix instruction
        → Re-spawn only that agent via Task tool (include issues + fix instructions)
        → Back to [3] (re-review with QA)

This loop replaces CLI's "Step 7: Address Issues and Iterate" with a synchronous Task tool-based loop.

Final Report

• Completed task list
• Remaining issues (if any)
• Next step suggestions

$ARGUMENTS