---

name: architect description: Hierarchical task execution with planning, verification, and replanning. Use for complex multi-step tasks with dependencies. Planner creates manifest, workers execute steps, inspector verifies.

The Architect: Hierarchical Task Execution Protocol

When to Activate

Use this skill when the user requests:

• Complex tasks with sequential dependencies
• Multi-step execution requiring verification
• Tasks where order matters
• Workflows that may need replanning
• User explicitly mentions "architect", "execute workflow", or "step-by-step task"

Architecture

Hierarchical Pyramid with Verification Loops

The Blueprint (Planning)

• Planner agent analyzes the task
• Creates numbered manifest of steps
• Identifies dependencies between steps
• Sets success criteria for each step

The Construction (Execution)

• Worker agent executes Step 1
• Reports back with results and status
• Inspector verifies success against criteria

The Inspection (Verification)

• If PASS: Hand out next step
• If FAIL: Trigger replanning
  • Planner reviews failure
  • Adjusts remaining steps
  • Worker retries or pivots

The Completion

• All steps verified
• Final summary of execution
• Artifacts collected

Invocation

Copypython3 ~/.claude/skills/architect/architect.py "Your complex task"

Options

Copypython3 ~/.claude/skills/architect/architect.py "Your task" --max-retries 3 --verbose

Example Usage

User: "Set up a new Python project with tests, CI, and documentation"

Copypython3 ~/.claude/skills/architect/architect.py "Create a new Python package called 'greenhydrogen' with pytest, GitHub Actions CI, and Sphinx documentation"

Execution Flow

Copy[Planner] → Blueprint (5 steps)
    ↓
[Worker] → Step 1 → [Inspector] ✓ Pass
    ↓
[Worker] → Step 2 → [Inspector] ✗ Fail
    ↓
[Planner] → Replan (adjust steps 3-5)
    ↓
[Worker] → Step 2 (retry) → [Inspector] ✓ Pass
    ↓
... continue until complete

Why This Works

1. Explicit Dependencies: Steps execute in correct order
2. Verification Gates: Each step validated before proceeding
3. Adaptive Replanning: Failures trigger intelligent adjustment
4. Audit Trail: Full log of what was attempted and why

Requirements

• Claude Code CLI installed and authenticated (claude command available)
• Python package: rich (for formatted output)