AgentSkillsCN

plan

采用原生 Claude 进行规划，并辅以架构师/评论家的并行自评

SKILL.md
--- frontmatter
name: plan
description: Claude-native planning with parallel architect/critic self-review
argument-hint: "[task description]"

Planning

Execute a multi-round planning session with architect/critic review.

Execution

  1. •Load protocol: Read agents/planner.md to load the full planner protocol. You execute it directly - do NOT spawn a planner agent.
  2. •Configure reviewers: Use coral:architect and coral:critic as reviewers (full review loop, up to 5 rounds). Only reviewers are spawned as subagents.
  3. •Execute protocol: Follow the planner protocol steps yourself (gather context, write plan, review loop until no CRITICAL/HIGH, completion)
  4. •Project validation: If project instructions define workflow rules (e.g., review gates, post-implementation steps), follow them. If validation fails, revise the plan to address the issues and re-validate until it passes.
  5. •Present plan: Show the final plan to the user

Context Enhancement

From the current conversation, identify and include:

  • •Task description and acceptance criteria
  • •File paths and code sections relevant to the plan
  • •Constraints or preferences stated by the user

Error Policy

If agents/planner.md cannot be read, report the error to the user.