orchestrator

Agent Orchestrator

You are activating the Agent Orchestrator workflow. Follow these steps:

1. Discover Available Agents

Scan for installed specialist agents:

ls ~/.claude/agents/*.md 2>/dev/null
ls .claude/agents/*.md 2>/dev/null

Read each agent's YAML frontmatter to understand their capabilities. Build a roster of available specialists before planning.

2. Gather Context

• Read STATUS.md if it exists in the project root — absorb project context silently
• Understand the user's feature request or task description

3. Draft an Implementation Plan

Create a structured plan with:

• Clear phases with agent assignments matched to installed agents
• Deliverables for each phase
• Dependency graph (which phases depend on which)
• Execution mode recommendation (sequential, parallel, or hybrid)

4. Present and Confirm

Always present the plan to the user before executing. If parallel phases are identified, explicitly ask for permission before running them simultaneously.

5. Execute

Spawn subagents with detailed, scoped prompts including:

• Role context and task scope
• Input artifacts from previous phases
• Output expectations and constraints
• Project conventions and decisions already made

6. Synthesize and Update

• Collect and synthesize results from all phases
• Update STATUS.md if the project uses it
• Offer to save the plan for reuse with /follow-plan