| name | intake |
| description | This skill should be used when starting a new coding task to understand the request, inventory capabilities, gather context, and classify task complexity. Use when the orchestrator invokes phase 0 or when analyzing a new task. |
Phase 0: Intake
Purpose
Understand the request, inventory capabilities, gather missing context, decide workflow path.
Process
Inventory available capabilities:
- Check available skills (what workflows/patterns are available?)
- Check MCP servers (what external integrations are connected?)
- Check project-specific tools (custom scripts, build systems, test runners)
- Search for deferred tools relevant to this task
- Note which capabilities are relevant to this task
Classify the request:
- Trivial (typo fix, one-line change) → Skip to implementation, no phases
- Simple (single-file, clear requirements) → Abbreviated workflow
- Complex (multi-file, unclear scope, architectural) → Full workflow
Gather context:
- Read relevant existing code before asking questions
- Check project CLAUDE.md for local conventions
- Review recent commits if task relates to recent work
- Check if existing skills apply (TDD, debugging, domain-specific)
Clarify requirements (if needed):
- One question at a time
- Multiple choice when possible
- Stop when you have: clear goal, success criteria, constraints
Produce intake summary: Write to docs/scratch/
/intake.md: - What: One sentence description
- Why: Purpose/motivation
- Success criteria: How we know it's done
- Constraints: Tech choices, patterns to follow, things to avoid
- Relevant capabilities: Skills, MCP servers, tools to use
- Workflow: Trivial/Simple/Complex
Checkpoint
Present intake summary to user. Get approval before Phase 1.
Escalation
- Request is ambiguous after 2-3 clarifying questions
- Request conflicts with existing architecture
- Required capability (MCP server, tool) is unavailable
- Request seems to require external dependencies not yet approved