4k Dimillian

orchestrate-batch-refactor Skill

通过并行的多智能体分析与实现,高效规划和执行大规模重构或重写工作。当用户要求重构大量文件、拆分工作流、分析目标代码区域,并协调具有明确职责和依赖感知执行的子智能体时使用。

安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。

查看源码

技能指令原文(SKILL.md)

Orchestrate Batch Refactor

Overview

Use this skill to run high-throughput refactors safely.
Analyze scope in parallel, synthesize a single plan, then execute independent work packets with sub-agents.

Inputs

  • Repo path and target scope (paths, modules, or feature area)
  • Goal type: refactor, rewrite, or hybrid
  • Constraints: behavior parity, API stability, deadlines, test requirements

When to Use Parallelization

  • Use this skill for medium/large scope touching many files or subsystems.
  • Skip multi-agent execution for tiny edits or highly coupled single-file work.

Core Workflow

  1. Define scope and success criteria.
  • List target paths/modules and non-goals.
  • State behavior constraints (for example: preserve external behavior).
  1. Run parallel analysis first.
  • Split target scope into analysis lanes.
  • Spawn explorer sub-agents in parallel to analyze each lane.
  • Ask each agent for: intent map, coupling risks, candidate work packets, required validations.
  1. Build one dependency-aware plan.
  • Merge explorer output into a single work graph.
  • Create work packets with clear file ownership and validation commands.
  • Sequence packets by dependency level; run only independent packets in parallel.
  1. Execute with worker agents.
  • Spawn one worker per independent packet.
  • Assign explicit ownership (files/responsibility).
  • Instruct every worker that they are not alone in the codebase and must ignore unrelated edits.
  1. Integrate and verify.
  • Review packet outputs, resolve overlaps, and run validation gates.
  • Run targeted tests per packet, then broader suite for integrated scope.
  1. Report and close.
  • Summarize packet outcomes, key refactors, conflicts resolved, and residual risks.

Work Packet Rules

  • One owner per file per execution wave.
  • No parallel edits on overlapping file sets.
  • Keep packet goals narrow and measurable.
  • Include explicit done criteria and required checks.
  • Prefer behavior-preserving refactors unless user explicitly requests behavior change.

Planning Contract

Every packet must include:

  1. Packet ID and objective.
  2. Owned files.
  3. Dependencies (none or packet IDs).
  4. Risks and invariants to preserve.
  5. Required checks.
  6. Integration notes for main thread.

Use references/work-packet-template.md for the exact shape.

Agent Prompting Contract

  • Use the prompt templates in references/agent-prompt-templates.md.
  • Explorer prompts focus on analysis and decomposition.
  • Worker prompts focus on implementation and validation with strict ownership boundaries.

Safety Guardrails

  • Do not start worker execution before plan synthesis is complete.
  • Do not parallelize across unresolved dependencies.
  • Do not claim completion if any required packet check fails.
  • Stop and re-plan when packet boundaries cause repeated merge conflicts.

Validation Strategy

Run in this order:

  1. Packet-level checks (fast and scoped).
  2. Cross-packet integration checks.
  3. Full project safety checks when scope is broad.

Prefer fast feedback loops, but never skip required behavior checks.