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

implement-task

Dispatches a planned task/feature to the task-executor agent for worktree-isolated TDD implementation

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 implement-task

技能说明


name: implement-task description: Dispatches a planned task/feature to the task-executor agent for worktree-isolated TDD implementation argument-hint: "[task-number-or-plan-path] [--project name]"

Task Implementation Skill

Dispatches a previously planned task or feature to the task-executor agent for worktree-isolated TDD implementation.

Explicit Project Input

If arguments include --project <name>, resolve that project directly from the repository's .groundwork.yml for this invocation. Treat it as authoritative; do not depend on or change persisted project selection. In runner mode, GROUNDWORK_PROJECT and GROUNDWORK_PROJECT_ROOT provide the same invocation-local selection.

Runner Mode

If session context contains GROUNDWORK_RUNNER_MODE=true:

  • Skip the model recommendation pre-flight and all questions.
  • In task mode, skip Step 2 entirely; the base checkout must remain clean. Tell the task-executor to record In Progress inside the task worktree and include it in the implementation commit.
  • Forward the runner-supplied absolute worktree path exactly; do not let the executor choose another directory.
  • Forward the runner-supplied task branch exactly; monorepos use a project-qualified branch to prevent overlapping task IDs from colliding.
  • When the runner supplies RESUME EXISTING WORKTREE=true, verify and reuse that exact registered worktree. Tell the executor to inspect existing commits, working state, tests, and plan progress; do not repeat completed implementation work.
  • Forward the runner receipt token exactly.
  • The runner exclusively owns task-worktree creation, reconciliation, cleanup, staging, and commits. The task-executor edits and tests the precreated worktree, then proposes an expressive commit subject and body.
  • Return the task-executor's versioned JSON RESULT: IMPLEMENTED receipt unchanged.

Token Discipline

This skill orchestrates implementation dispatch. Every turn re-reads the full context window, so unnecessary turns are expensive.

  1. No narration turns. Do not output text-only turns like "Let me dispatch the task" or "Now I'll launch the executor." Combine text with a tool call in the same turn, or skip the text entirely.
  2. Batch tool calls. When multiple tool calls are independent, issue them all in one turn.
  3. No waiting updates. Do not output "Waiting for results..." turns. Wait silently until results arrive.
  4. Keep context lean. Do not read file contents you won't use directly. The task-executor agent has Read/Grep/Glob and will read files in its own context window.

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": "Implementation benefits from consistent multi-domain reasoning.\n\nTo switch: cancel, run `/effort high` (and `/model sonnet` if on Haiku), then re-invoke.",
    "header": "Effort check",
    "options": [
      { "label": "Continue" },
      { "label": "Cancel — I'll switch first" }
    ],
    "multiSelect": false
  }]
}

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

Step 0: Resolve Project Context

Before dispatching, 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. Proceed with the resolved project context.

Step 1: Resolve Plan File

Parse input from the caller's conversation context. Three modes:

  • plan_file_path provided → Read the plan file header to extract Identifier, Mode, Branch prefix, Tasks path.
  • task_id only (no plan_file_path) → Derive path .groundwork-plans/TASK-NNN-plan.md. If file does not exist, output RESULT: FAILURE | No plan found for TASK-NNN. Run plan-task first. and stop.
  • task_id + plan_file_path → Use the provided plan file path. Verify it exists.
  • Neither → Output RESULT: FAILURE | No plan_file_path or task_id provided. Run plan-task first. and stop.

Read ONLY the ## Context header from the plan file to extract:

  • identifier (e.g., TASK-004 or FEATURE-user-login)
  • mode (task, feature, or task+blurb)
  • branch_prefix (task or feature)
  • tasks_path (path to tasks file, or N/A)
  • specs_dir

Do NOT read the full plan content into this orchestrator's context. The task-executor agent will read it.

Step 2: Update Task Status (task mode only)

Skip this step if identifier starts with FEATURE-.

Also skip this step if GROUNDWORK_RUNNER_MODE=true. Add the equivalent status update to the task-executor instructions so it occurs and is committed inside the task worktree, never in the base checkout.

Update the task file to **Status:** In Progress and update the status table in {{specs_dir}}/tasks/_index.md or {{specs_dir}}/tasks.md (change the task's row to In Progress).

Step 3: Dispatch to task-executor

Dispatch to the task-executor agent with a fresh context window. This agent has use-git-worktree and test-driven-development skills preloaded — it does not need to call Skill() or spawn subagents.

Build the Agent prompt — pass paths, not content.

Task mode (identifier starts with TASK-)

Agent(
  subagent_type="groundwork:task-
阅读完整来源 ↗