Analysis Phase Skill

ROOT AGENT ONLY - Invokes skills in parallel, runs only from root Claude Code agent.

Purpose: Gather context for routing and planning decisions Trigger: First phase of workflow commands

Command-Specific Checks

Each command specifies which checks to run in parallel:

IMPORTANT: Refer to the command file for the exact checks to spawn.

Available Check Skills

How to Execute

Spawn check agents in PARALLEL using dedicated analysis agents (zero tool access):

# Example for /ms or /build
Task 1: subagent_type="core-claude-plugin:generic:scope-check-agent"
  Prompt: "Analyze this prompt and return scope (CWD: {CWD}): {USER_PROMPT}"

Results are returned inline. Each Task call returns the agent's output directly in the response. Read the result from the Task response and proceed immediately.

Why dedicated agents? These agents specify tools: TodoWrite (a minimal core tool) instead of inheriting all tools, which eliminates ~37k tokens of MCP tool schemas per agent.

Collect results from the Task responses based on which checks were run.

Skill Outputs

Each agent invokes its corresponding skill for the detailed algorithm:

See individual skill files for detailed algorithms and examples.

Output Example

{
  "scope": {
    "targets": ["/home/user/code/metasaver-com"],
    "references": ["/home/user/code/rugby-crm"]
  }
}

Next Phase

After analysis, proceed to requirements-phase.

Integration

Called by:

• /audit command
• /build command
• /ms command

Calls:

• scope-check skill (returns {targets: string[], references: string[]})

Next step: requirements-phase

Example

Command: /build "Add Applications screen to metasaver-com, look at rugby-crm for patterns"

Analysis Phase:
  → scope-check: { targets: ["/home/user/code/metasaver-com"], references: ["/home/user/code/rugby-crm"] }

Output:
  {
    scope: {
      targets: ["/home/user/code/metasaver-com"],
      references: ["/home/user/code/rugby-crm"]
    }
  }

Next: requirements-phase