deep-plan

Deep Plan

Multi-perspective architecture planning. Spawns specialist agents for different quality-attribute concerns that evaluate trade-offs before you write code.

When This Skill Activates

Trigger on architecture and planning requests:

• User says: "plan", "architect", "design", "how should I architect..."
• User asks: "deep plan", "plan the architecture", "technical design"
• User wants pre-implementation architecture evaluation

Do NOT activate on every task. This is for deliberate architecture planning, not quick feature additions. Simple features don't need a 6-agent swarm.

Phase 1: Context Gathering

Before spawning agents, establish scope:

1. Check for deep-research artifact: Scan recent conversation context for the structured JSON artifact from deep-research (contains task, tech_stack, approaches, findings, verdict). Validate the artifact has at minimum: task, tech_stack, and at least one approach with a summary. If any required field is missing or empty, treat as if no artifact was found and fall through to step 2. If valid, use it as primary input — skip step 2 and use the recommended approach as the starting architecture to evaluate.
2. If no artifact: Gather context manually:
   • Run git rev-parse --show-toplevel for {repo_root}
   • Check for package manifests to identify {tech_stack}. If no manifests are found, infer tech_stack from primary file extensions (.py, .ts, .go, etc.) or ask the user. Never pass an empty tech_stack to planning agents.
   • Ask the user what they're building and what constraints matter most
   • Read existing code in the affected area to understand current architecture
3. speak-memory: If .speak-memory/index.md exists and an active story matches, read it for context. Skip if .speak-memory/ does not exist.
4. Frame the architecture: State what is being planned:
   • Feature/System: What is being designed
   • Scope: Which components, modules, or layers are affected
   • Constraints: Performance targets, security requirements, compatibility needs
   • Starting approach: From deep-research artifact, or from user description
5. Set scope budget: "Budget: 5 planning specialists + 1 trade-off arbiter = 6 agent calls." Present to user: