ralph-swarm:go

ralph-swarm:go

Resume swarm execution after the planning-review pause. This command picks up where /ralph-swarm:start left off.

Step 0: Read Project Context

1. Check if CLAUDE.md exists in the project root. If it does, read it with the Read tool. Every agent you delegate to must receive the CLAUDE.md content.

Step 1: Load and Validate State

1. Read .ralph-swarm-state.json from the project root using the Read tool.
2. If the file does not exist, output this error and stop:

   Error: No active swarm found. No .ralph-swarm-state.json in project root.
   Run /ralph-swarm:start "your goal" to begin.
   

3. Parse the JSON. Extract phase, name, goal, specPath, flags, and execution fields.
4. Verify that phase is exactly "planning-review" or "planning-complete". If it is anything else, output this error and stop:

   Error: No plan awaiting review. Current phase is "<phase>".
   - If phase is "planning": planning is still in progress.
   - If phase is "execution": execution is already running. Use /ralph-swarm:status to check progress.
   - If phase is missing: state file may be corrupted. Run /ralph-swarm:cancel and start over.
   

Step 2: Read Tasks

1. Read the tasks file at <specPath>/tasks.md using the Read tool.
2. If the file does not exist or is empty, output this error and stop:

   Error: Tasks file not found at <specPath>/tasks.md. The planning phase may have failed.
   Check the spec files at <specPath>/ and run /ralph-swarm:cancel to reset.
   

3. Parse the tasks file to extract all tasks. The format is defined in ../start/task-format.md. Each task has:
   • id: TASK-NNN identifier (e.g., TASK-001, TASK-002)
   • title: the feature slice title
   • complexity: S, M, or L
   • files: list of files to CREATE or MODIFY (with operation type)
   • dependencies: list of TASK-IDs this task depends on (or "None")
   • description: end-to-end slice description
   • context to read: files and design sections the executor needs
   • verification: exact command to verify completion
4. Parse the File Manifest table at the bottom of tasks.md. This provides a quick-scan summary of which tasks touch which files.
5. Count the total number of tasks.

Step 3: Update State for Execution

Update .ralph-swarm-state.json with these changes:

• Set phase to "execution"
• Set pausedAfter to null
• Set execution.totalTasks to the total task count
• Set execution.taskIndex to 0
• Set execution.completedTasks to []
• Set execution.failedTasks to []
• Set execution.iteration to 0
• Set execution.swarm to the value of flags.swarm
• Populate execution.tasks as an array of objects:

  [
    {"id": 1, "title": "task title", "status": "pending", "dependsOn": []},
    {"id": 2, "title": "task title", "status": "pending", "dependsOn": [1]},
    ...
  ]
  

Write the updated state file.

Step 4: Display Execution Start

Output to the user:

Starting execution for: <name>
Goal: <goal>
Mode: <sequential | swarm (parallel)>
Tasks: <totalTasks> total
Max iterations: <maxIterations>
Commit after tasks: <yes | no>

Step 5: Begin Execution

See ../start/execution-protocol.md for the full execution protocol covering both sequential and swarm modes.

Error Handling

• If the state file is malformed JSON, report the parse error and suggest running /ralph-swarm:cancel to reset.
• If tasks.md is malformed (cannot parse task structure), report what went wrong and suggest the user edit the file manually then re-run /ralph-swarm:go.
• If a teammate fails to start in swarm mode, reduce the teammate count and continue with available agents.
• Never silently swallow errors. Always report and update the state file.