AgentSkillsCN

lisa-yolo

以全自动模式继续推进史诗故事。

SKILL.md
--- frontmatter
name: lisa-yolo
description: Continue epic in full auto mode

Yolo Mode (Full Auto)

Epic name: Provided via arguments

Full automatic execution with no checkpoints or confirmations.

YOLO RULES - FOLLOW STRICTLY

  1. NEVER stop to summarize - No progress updates. Just work.
  2. NEVER ask for confirmation - No "Ready to continue?". Just proceed.
  3. NEVER explain what you're doing - Don't narrate. Execute.
  4. ALWAYS execute next phase immediately - After phase completes, start next phase.
  5. ONLY stop when truly done - All tasks done OR all remaining blocked.
  6. Maximize progress per response - Execute as many phases/tasks as possible.
  7. ALWAYS use fresh agents - Each phase uses Task tool with clean context.
  8. ALWAYS verify completion - Use verify_* tools after each phase.
  9. RETRY on failure - Up to 3 attempts per phase before marking blocked.

Yolo is for autonomous execution. User has walked away. Every stop wastes time.

If unsure, keep working. Better to complete extra work than stop and ask.


Verify Spec Exists

If no spec:

code
No spec found for epic '<name>'.

Create the epic first: /lisa-create-epic <name>

STOP.


Activate/Resume Yolo in .state

Read current .state to check if yolo was previously active.

If yolo.active: true already exists:

code
Yolo mode was previously active (iteration ${currentIteration}).

Resuming from current state:
  ✓ Spec - ${specComplete ? 'complete' : 'not started'}
  ${researchComplete ? '✓' : '○'} Research - ${researchComplete ? 'complete' : 'not started'}
  ${planComplete ? '✓' : '○'} Plan - ${planComplete ? 'complete' : 'not started'}
  ● Execute - ${completedTasks}/${totalTasks} tasks done

Continuing yolo execution...

Increment iteration counter. Check if iteration > maxIterations:

code
Error: Max yolo iterations (${maxIterations}) reached.

The epic may be stuck or too complex for fully automatic execution.

Resume with:
  • /lisa-continue <name> - Switch to interactive mode with checkpoints

STOP.

If yolo not active or new start: Update .state:

json
{
  "yolo": {
    "active": true,
    "iteration": 1,
    "maxIterations": 100,
    "startedAt": "<ISO timestamp>"
  }
}

Execute All Remaining Phases

Run phases WITHOUT asking, using fresh sub-agents with verification and retry:

1. Research (if not done)

DO NOT ask. Just do it.

ATTEMPT 1:

  1. Call build_research_context("${epicName}")
  2. Call Task tool with:
    • prompt: returned research prompt
    • subagent_type: "explore"
  3. Call verify_research_complete("${epicName}")
  4. If verification.complete === true: Immediately proceed to planning

ATTEMPT 2 (if incomplete):

  1. Call Task tool with STRONGER prompt:
    code
    [YOLO RESEARCH RETRY - MUST COMPLETE NOW]
    
    Previous attempt INCOMPLETE. Issues: [verification.issues]
    
    YOLO MODE = NO STOPPING. COMPLETE IMMEDIATELY:
    1. Read spec.md
    2. Write complete research.md
    3. Update .state
    
    DO NOT stop. DO NOT explain. JUST DO IT.
    
    [original research prompt]
    
  2. Call verify_research_complete("${epicName}")
  3. If complete: Proceed to planning

ATTEMPT 3 (if still incomplete):

  1. Final attempt with URGENT prompt
  2. Call verification
  3. If still incomplete:
    • Update .state.yolo.active = false
    • Show: "Yolo mode stopped: Research failed after 3 attempts. Issues: [list]. Use /lisa-continue ${epicName} for interactive mode."
    • STOP

2. Plan (if not done)

DO NOT ask. Just do it.

Same retry pattern as Research:

ATTEMPT 1:

  1. Call build_planning_context("${epicName}")
  2. Call Task tool with:
    • prompt: returned planning prompt
    • subagent_type: "general"
  3. Call verify_planning_complete("${epicName}")
  4. If complete: Immediately proceed to execution

ATTEMPTS 2-3: Same retry logic with stronger prompts

If still incomplete after 3 attempts:

  • Stop yolo mode
  • Report failure
  • STOP

3. Execute

Use execution.maxParallelTasks from config (default: 1).

TASK EXECUTION WITH VERIFICATION:

Loop until all done or blocked:

  1. Call get_available_tasks(epicName)

  2. For each available task (up to maxParallelTasks):

    ATTEMPT 1:

    • Call build_task_context(epicName, taskId)
    • Call Task tool with prompt
    • Call verify_task_complete(epicName, taskId)
    • If complete: Move to next task

    ATTEMPT 2 (if incomplete):

    • Call Task tool with STRONGER prompt:
      code
      [YOLO TASK RETRY - COMPLETE IMMEDIATELY]
      
      Task ${taskId} NOT COMPLETE. Issues: [verification.issues]
      
      YOLO RULES:
      - DO NOT stop
      - DO NOT summarize
      - COMPLETE THE TASK NOW
      - Update status to "done"
      - Add full ## Report section
      
      [original task prompt]
      
    • Call verification
    • If complete: Move to next task

    ATTEMPT 3 (if still incomplete):

    • Final attempt with URGENT prompt
    • Call verification
    • If still incomplete:
      • Mark task as blocked in task file
      • Continue with other tasks (DO NOT stop yolo)
  3. Call get_available_tasks again

  4. REPEAT - No summaries, just keep executing

If all remaining tasks are blocked:

  • Update .state.yolo.active = false
  • Show: "Yolo mode stopped: All remaining tasks blocked after retries."
  • List blocked tasks and reasons
  • STOP

Completion

On all tasks done:

  • Update .state (executeComplete: true, yolo.active: false)
  • Handle git based on git.completionMode config
  • Show final summary:
    code
    Epic complete! All X tasks finished.
    
    Changes: [brief summary]
    

On tasks blocked:

  • Update .state (yolo.active: false)
  • Report which tasks blocked and why