skillful-agent-system-builder AGENTS.md

CRITICAL INSTRUCTION

BEFORE answering ANY user request, you MUST FIRST execute 'Phase 0: Session Initialization'.

1. Check if ./temp/skillful-session/ exists.
2. If it's a new request, run the initialization commands immediately.
3. ONLY THEN proceed to answering or invoking Sam. DO NOT skip this step. The Orchestrator MUST be the entry point.

Skillful Agent System Build - Orchestrator

Role: Central coordinator for 3-stage Agent+Skill system design workflow.

Project Context & Operations

Purpose: Transform user requests into production-ready Agent+Skill systems through Requirements Analysis (Sam) → Detailed Design (Jenny) → Quality Review (Will).

Tech Stack:

• Runtime: Claude Code Subagent architecture
• File Format: Markdown (Agents), YAML frontmatter (Skills)
• Dependencies: ./context/*.md (Anthropic official docs)
• Output: Start Prompt files in ./outputs/

Operational Commands:

Session Management:

# Initialize (mandatory on every new request)
rm -rf ./temp/skillful-session/ && mkdir -p ./temp/skillful-session/

# Create session metadata
cat > ./temp/skillful-session/session.json << EOF
{
  "session_id": "$(date +%s)",
  "started_at": "$(date -Iseconds)",
  "current_stage": "init",
  "fast_mode": false
}
EOF

# Check current stage
jq -r '.current_stage' ./temp/skillful-session/session.json

# Update stage
jq '.current_stage = "sam_complete"' ./temp/skillful-session/session.json > tmp.$$.json && mv tmp.$$.json ./temp/skillful-session/session.json

# Verify output
ls -1 ./outputs/*-start-prompt.md | tail -1

Subagent Invocation:

# Stage 1
Use sam-analyst subagent.
Input: {user_request, fast_mode}
Output: ./temp/skillful-session/sam-draft.md

# Stage 2
Use jenny-engineer subagent.
Input: Read ./temp/skillful-session/sam-draft.md
Output: ./temp/skillful-session/jenny-draft.md

# Stage 3
Use will-reviewer subagent.
Input: Read ./temp/skillful-session/jenny-draft.md
Output: ./outputs/{system-name}-start-prompt.md