跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
浏览器与自动化

just-do-it-swarming

Execute all remaining tasks using agent teams for context isolation. Each task runs in its own session. Usage /groundwork:just-do-it-swarming [--parallel]

etr · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add etr/groundwork --skill just-do-it-swarming

技能说明


name: just-do-it-swarming description: Execute all remaining tasks using agent teams for context isolation. Each task runs in its own session. Usage /groundwork:just-do-it-swarming [--parallel] allowed-tools: ["Read", "Edit", "Write", "Bash", "Glob", "Grep", "Task", "AskUserQuestion", "Skill"] disable-model-invocation: true

Swarming Batch Task Execution

Executes all remaining tasks using agent teams for context isolation. Each task runs in a separate teammate session with its own context window. Teammates are full Claude Code sessions that can spawn subagents (Plan, task-executor, 9 validators), solving the context accumulation problem of inline batch execution.

When to use this vs /groundwork:just-do-it:

  • Use this (/groundwork:just-do-it-swarming) for large batches (>5 tasks) where context accumulation is a concern
  • Use /groundwork:just-do-it for small batches or when agent teams are not available

Requirement: Agent teams must be enabled (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1" in settings.json env section).

Model: Teammates spawn with model: "opus" for extended context.

Pre-flight: Model Recommendation

Your current effort level is {{effort_level}}.

Skip this step silently if effort is high, xhigh, or max (the scale is low < medium < high < xhigh < max, so xhigh and max are already above high) AND you are Sonnet or Opus. If effort is low or medium (i.e. below high), you MUST show the recommendation prompt — regardless of model. If you are not Sonnet or Opus, you MUST show the recommendation prompt - regardless of effort level.

Otherwise → use AskUserQuestion:

{
  "questions": [{
    "question": "Do you want to switch? Team orchestration with worktree coordination across agents benefits from consistent reasoning.\n\nTo switch: cancel, run `/effort high` (and `/model sonnet` if on Haiku), then re-invoke this skill.",
    "header": "Recommended: Sonnet or Opus at high effort",
    "options": [
      { "label": "Continue" },
      { "label": "Cancel — I'll switch first" }
    ],
    "multiSelect": false
  }]
}

If the user selects "Cancel — I'll switch first": output the switching commands above and stop. Do not proceed with the skill.

Workflow

Step 0: Resolve Project Context

Before loading tasks, ensure project context is resolved:

  1. Monorepo check: Does .groundwork.yml exist at the repo root?
    • If yes → Is {{project_name}} non-empty?
      • If empty → Invoke Skill(skill="groundwork:select-project") to select a project, then restart this skill.
      • If set → Project is {{project_name}}, specs at {{specs_dir}}/.
    • If no → Continue (single-project repo).
  2. CWD mismatch check (monorepo only):
    • Skip if not in monorepo mode or if the project was just selected in item 1 above.
    • If CWD is the repo root → fine, proceed.
    • Check which project's path CWD falls inside (compare against all projects in .groundwork.yml).
    • If CWD is inside the selected project's path → fine, proceed.
    • If CWD is inside a different project's path → warn via AskUserQuestion:

      "You're working from <cwd> (inside [cwd-project]), but the selected Groundwork project is [selected-project] ([selected-project-path]/). What would you like to do?"

      • "Switch to [cwd-project]"
      • "Stay with [selected-project]" If the user switches, invoke Skill(skill="groundwork:select-project").
    • If CWD doesn't match any project → proceed without warning (shared directory).
  3. Proceed with the resolved project context. All {{specs_dir}}/ paths will resolve to the correct location.

Step 1: Load and Analyze Tasks

  1. Read the tasks file to find all tasks:

    • Single file: {{specs_dir}}/tasks.md
    • Directory: {{specs_dir}}/tasks/ (aggregated in sorted order)
  2. Parse all tasks and extract:

    • Task ID (e.g., TASK-001)
    • Title
    • Status (Not Started, In Progress, Complete, Blocked)
    • Dependencies (Blocked by: field)
  3. Build dependency graph and calculate execution order:

    • Topological sort respecting dependencies
    • Blocked tasks cannot execute until dependencies complete

Detection: Check for file first (takes precedence), then directory.

Step 2: Present Summary and Confirm

Present a summary to the user:

Swarming Batch Execution Summary

Mode: Agent Teams (swarming) Total tasks: X Already complete: Y Remaining: Z

Context Isolation

Each task runs in a separate teammate session with its own context window:

  • Teammates are full Claude Code sessions with the complete system prompt
  • Each teammate can spawn subagents (Plan, task-executor, 9 validation agents)
  • Context is fully isolated between tasks — no accumulation in the lead's conversation
  • Changes merged after each task completes

Worktree Isolation

Each task also executes in an isolated git worktree:

  • Branch: task/TASK-NNN created from current HEAD
  • Working directory: .worktrees/TASK-NNN
  • Changes merged automatically after each task completes successfully
  • Worktrees cleaned up after successful merge

Execution Order

  1. TASK-NNN: [Title]
  2. TASK-NNN: [Title] ...

Blocked Tasks (will execute after dependencies complete)

  • TASK-NNN: [Title] (blocked by TASK-XXX)

Ask for confirmation before proceeding.

If user declines, stop and suggest alternatives:

  • /groundwork:just-do-it for inline execution (no agent teams required)
  • /groundwork:work-on N to work on a specific task
  • /groundwork:work-on-next-task to work on just the next available task

Step 2.5: Validate Agent Teams Enabled

Before creating the team, verify agent teams are available by checking if the team creation tools are accessible.

If agent teams are not enabled:

"Agent teams are required for swarming mode but are not enabled. To enable them, add this to your settings.json:

{
  "env": {
阅读完整来源 ↗