AgentSkillsCN

arc-implementing

当您需要在工作树中统筹大型项目实施时,可使用此功能。

SKILL.md
--- frontmatter
name: arc-implementing
description: Use when orchestrating large project implementation in a worktree

arc-implementing

Overview

Orchestrator for large projects. Automatically expands epic → features → tasks → execution.

Use when:

  • project has dag.yaml
  • in a worktree session
  • epic.md or features/*.md present

Do not use when:

  • small projects (use writing-tasks + agent-driven directly)
  • tasks without a structured spec

Role

Implementer is the Orchestrator. It calls other skills and does not write code itself.

Trigger

  • in a worktree session
  • .arcforge-epic marker exists
  • or epic.md / features/*.md present

The Process

  1. For each epic in the worktree, run the following phases in order.
  2. Phase 0: Sync and check dependencies.
    • Run arc-coordinating to sync from base and check blocked_by.
    • If blocked_by is not empty: STOP and use the blocked format.
    • If ready: continue to Phase 1.
  3. Phase 1: Epic → Features.
    • Call arc-writing-tasks
    • Input: epic.md
    • Output: features list (may already exist in features/*.md)
  4. Phase 2: Per Feature.
    • 2a: Feature → Tasks.
      • Call arc-writing-tasks
      • Input: feature.md
      • Output: docs/tasks/<feature>-tasks.md
      • Quality gate: If tasks are vague or missing tests/commands, STOP and re-run arc-writing-tasks to refine.
    • 2b: Execute Tasks.
      • Call arc-agent-driven
      • Input: tasks file
      • Output: completed code + commits
  5. Phase 3: Feature complete.
    • Move to next feature, or finish the epic.

Skills Called

PhaseSkillInputOutput
0arc-coordinatingworktreesync + blocked status
1arc-writing-tasksepic.mdfeatures breakdown
2aarc-writing-tasksfeature.mdtasks file
2barc-agent-driventasks filecompleted code
2barc-dispatching-parallel(if review finds multiple issues)parallel fixes
Endarc-finishing-epiccompleted epicmerge decision

What Implementer Does NOT Do

  • ❌ Write code directly (delegate to agent-driven)
  • ❌ Split tasks manually (delegate to writing-tasks)
  • ❌ Perform reviews (handled inside agent-driven)
  • ❌ Run TDD cycle (handled inside agent-driven)

Completion Format

code
─────────────────────────────────────────────────
✅ Epic complete: <epic-name>

Features implemented:
- feature-1: 4 tasks, all passing
- feature-2: 6 tasks, all passing
- feature-3: 3 tasks, all passing

Total: 13 tasks, 0 failures
Commits: 13

Next: Use arc-finishing-epic to decide merge/PR/keep/discard
─────────────────────────────────────────────────

Blocked Format

Dependencies Not Ready

code
─────────────────────────────────────────────────
⚠️ Implementer blocked: waiting for dependencies

Epic: <epic-name>
Blocked by: <dep-1>, <dep-2>

To resolve:
1. Complete blocking epics first
2. Run `arc-coordinating` sync from base
3. Verify `blocked_by` is empty

Then resume implementer
─────────────────────────────────────────────────

Task Failure

code
─────────────────────────────────────────────────
⚠️ Implementer blocked

Epic: <epic-name>
Feature: <feature-name>
Task: <task-id>
Issue: [description]

To resolve:
1. [action]

Then resume implementer
─────────────────────────────────────────────────

Integration

  • Before: arc-coordinating (creates worktrees), arc-planning (creates DAG)
  • Uses: arc-writing-tasks, arc-agent-driven, arc-dispatching-parallel
  • After: arc-finishing-epic