Mycelium Plan

Update session state - Write invocation_mode: "single" to [state.json][session-state-docs]

Parse input:

• If user provided task description: Use it
• If empty: Ask user for task description

Provide context:

• Read [state.json][session-state-docs]
• Read .mycelium/context/*.md if exists (product, tech-stack, workflow)
• Read CLAUDE.md if exists

Load cached capabilities:

• Read discovered_capabilities from state.json
• If missing: error and suggest running /mycelium-context-load first
• Available capabilities cached from Phase 0 (Context Loading):
  • Skills (with fully-qualified names like mycelium:mycelium-work)
  • Agents (built-in and plugin agents)
  • MCP tools (if configured)

Analyze the task:

• Clarify requirements if ambiguous (use AskUserQuestion)
• Search codebase for relevant files (grep, glob)
• Break down into tasks with dependencies

Create detailed plan following the guidance below:

• Transform high-level requirements into actionable, parallel-executable tasks
• Apply proper task decomposition and dependency management
• Assign capabilities to each task (agent, skills, model)
• Verify assignments against cached capabilities
• Create plans that enable compound engineering where each task builds knowledge

Save plan to .mycelium/plans/YYYY-MM-DD-{track-id}.md using the [plan template][plan-template]. The frontmatter must conform to the [plan frontmatter schema][plan-schema].

Register plan in session state:

• Read state.json
• If a plan in plans[] has status: "in_progress", set it to "paused" (both in plans[] AND in that plan file's YAML frontmatter status field)
• Append new plan entry to plans[]:

  {
    "track_id": "{track_id}",
    "plan_file": "YYYY-MM-DD-{track-id}.md",
    "status": "in_progress",
    "created": "{timestamp}",
    "total_tasks": {count},
    "completed_tasks": 0
  }
  

• Set current_track to the new plan's { "id": "{track_id}", "type": "{type}", "plan_file": "..." }

Hand off to next phase: