Code-Index-MCP: Command for Claude Code

Execute Swim Lane with Parallel Sub-Agents

Inputs

• $1 = Phase document path (e.g., plans/phase-2-ingestion.md)
• $2 = Swim lane identifier (e.g., SL-1, lane-api-routes, etc.)

Execution Strategy

1. Load Phase Plan & Extract Swim Lane

Read the phase implementation document:

• @$1

Locate swim lane $2 within the document. Extract:

• Swim lane scope: Component boundaries, files, interfaces owned
• Task list: Individual work items with their dependencies
• Interface contracts: Frozen interfaces this lane depends on or provides
• Acceptance criteria: Definition of done for this lane

2. Verify Preconditions

Before spawning agents, check:

Interface Dependencies:

• All interfaces this lane consumes must be frozen (have IF-* gate marked complete)
• If any dependency is not ready, list them and STOP with message:

  ⚠️ Cannot proceed - waiting on interface freeze gates:
  - IF-0-xyz: [description]
  
  Recommend: Execute dependent lanes first or request interface definition

Git State:

• Current branch: !git branch --show-current
• Verify branch matches expected feature branch for this phase
• Check for uncommitted changes: !git status --short
• If dirty, warn but allow continuation (agents will commit their changes)

File Conflicts:

• Check if any files in this lane's scope are already modified in working tree
• List conflicts and ask user to confirm continuation

3. Task Dependency Analysis

From the task list for lane $2:

• Build dependency graph (which tasks can run in parallel vs sequential)
• Identify tasks with no dependencies ("ready" tasks)
• Group related tasks that should execute sequentially in one agent

Grouping heuristics:

• Tasks touching the same file → same agent (sequential)
• Tasks with direct data dependencies → same agent (sequential)
• Tasks on different files with no dependencies → parallel agents
• Avoid creating more than 5 parallel agents (diminishing returns)