AgentSkillsCN

agents-bootstrap

根据用户目标生成项目专属的 AGENTS.md 文件，并在覆盖前进行确认。

SKILL.md
--- frontmatter
name: agents-bootstrap
description: Generate a project-specific AGENTS.md from a user goal, then confirm before overwriting.
metadata:
  short-description: Bootstrap AGENTS.md for a real project

AGENTS Bootstrap

Generate a concise, project-specific AGENTS.md tailored to the user's goal.

Core rules

  • •Use assets/agents-template.md as the base structure.
  • •No placeholders in the final AGENTS.md.
  • •Keep the file concise and enforceable.
  • •Draft in chat first; ask for confirmation before writing.
  • •Before overwriting, create a backup: AGENTS.md.bak.<timestamp>.
  • •If MCP tools are in scope and docs/mcp-tools.md is missing or stale, instruct to use the mcp-tools-catalog skill before finalizing AGENTS.md.

Required inputs

Collect these if not provided:

  • •Role and objective
  • •Non-negotiable constraints
  • •Stack and data sources
  • •Testing toolchain (unit/integration/e2e commands, frameworks, MCP tools)
  • •Whether Issue CSV workflow is in scope

Ask at most 2 clarification questions, then proceed with stated assumptions. If the task continues across turns, re-invoke this skill to keep the rules active.

Output requirements

The final AGENTS.md must include (when applicable):

  • •Role & objective
  • •Constraints
  • •Tech & data sources
  • •Project testing strategy (tools + commands)
  • •E2E loop (plan → issues → implement → test → review → commit → regression)
  • •Plan & issue generation reference (use plan skill)
  • •Issue CSV policy (if Issue CSV workflow is in scope)
  • •Tool usage (MCP usage guidance)
  • •Testing policy reference (docs/testing-policy.md)
  • •Safety and output style

Write flow

  1. •Draft the tailored AGENTS.md in chat.
  2. •Ask: "Reply CONFIRM to overwrite AGENTS.md."
  3. •On confirmation:
    • •Backup current file to AGENTS.md.bak.<timestamp> if it exists.
    • •Write the new AGENTS.md.

Do not edit other files.