360 Secure SkillHub
360 安全技能中心

agent-rules

扫描项目技术栈、构建/CI 配置与 git 历史，按模板生成根级与分作用域 AGENTS.md，并可检测结构、命令、内容新鲜度与质量评分，同时创建 CLAUDE.md/GEMINI.md 兼容链接。

分类：开发
netresearch
热度 0.00
最近更新 2026-10-24 03:08:37

安装

发给 Agent
帮我安装这个技能：npx secure-skills-cli add netresearch/agent-rules-skill/agent-rules
兼容性
Claude Code 专属
联网要求
不联网
数据出网
数据不出本机
Metadata
name
agent-rules
description
Use when creating or updating AGENTS.md files, .github/copilot-instructions.md, or other AI agent rule files, onboarding AI agents to a project, standardizing agent documentation, or when anyone mentions AGENTS.md, agent rules, project onboarding, or codebase documentation for AI agents.
license
(MIT AND CC-BY-SA-4.0). See LICENSE-MIT and LICENSE-CC-BY-SA-4.0
compatibility
Requires bash 4.3+, jq 1.7+, git 2.0+.
metadata
{"author":"Netresearch DTT GmbH","version":"3.16.5","repository":"https://github.com/netresearch/agent-rules-skill"}
allowed-tools
Bash(${CLAUDE_SKILL_DIR}/scripts/*) Bash(bash ${CLAUDE_SKILL_DIR}/scripts/*) Bash(git:*) Bash(jq:*) Bash(grep:*) Bash(find:*) Read Glob Grep

AGENTS.md Generator Skill

Generate and maintain AGENTS.md files following the agents.md convention. AGENTS.md is FOR AGENTS, not humans.

When to Use

  • Creating or updating AGENTS.md for new/existing projects
  • Scaffolding a new repository — ship AGENTS.md with the initial commits; retrofitting later needs full re-verification
  • Standardizing agent documentation across repositories
  • Checking AGENTS.md freshness after code changes
  • Onboarding AI agents to an unfamiliar codebase

Scripts

Call every script by its full path: bash ${CLAUDE_SKILL_DIR}/scripts/<name> PATH. Calling one relative to the working directory is not covered by the frontmatter rule and raises a permission prompt per call.

Script Purpose
generate-agents.sh PATH Generate AGENTS.md files
validate-structure.sh PATH Validate structure compliance
check-freshness.sh PATH Check if files are outdated
verify-content.sh PATH Verify documented files/commands match codebase
verify-commands.sh PATH Verify documented commands execute
score-agents.sh PATH Grade AGENTS.md quality, worst-first
detect-project.sh PATH Detect language, version, build tools
detect-scopes.sh PATH Identify directories needing scoped files
extract-commands.sh PATH Extract commands from build configs
extract-ci-rules.sh PATH Extract CI quality gates and version matrix
extract-architecture-rules.sh PATH Extract module boundaries
extract-adrs.sh PATH Extract architectural decision records
extract-github-rulesets.sh PATH Extract GitHub rulesets and merge rules

See references/scripts-guide.md for full options.

Workflow

  1. Detect: detect-project.sh + detect-scopes.sh — stacks and subsystems
  2. Extract: extract-commands.sh, extract-ci-rules.sh — gather facts
  3. Generate: generate-agents.sh --style=thin (default) or --verbose
  4. Verify: verify-content.sh + verify-commands.sh -- MANDATORY before done
  5. Line cap: before committing any root AGENTS.md edit, wc -l AGENTS.md must be under 150 (harness AH-02); run the repo's verify-harness.sh where it exists -- CI's harness check fails otherwise

--update preserves curated content outside <!-- GENERATED --> markers.

Core Principles

  • Structured over Prose -- tables parse faster than paragraphs
  • Never Fabricate -- only document what exists; verify every command and path
  • Pointer Principle -- point to files, don't duplicate content
  • Auto Symlinks -- CLAUDE.md/GEMINI.md by default (ai-tool-compatibility.md)

References

File Contents
verification-guide.md Verification steps, anti-bloat, preservation check
fleet-sync-sweep.md Fleet-wide AGENTS.md sweeps
scripts-guide.md Script options, validation checklist
quality-rubric.md Grading rubric
ai-tool-compatibility.md 16-agent compatibility matrix
output-structure.md Root/scoped sections
git-hooks-setup.md Hook framework setup
examples/ Complete examples
ai-contribution-guidelines.md "3 Cs" AI-contribution framework
directory-coverage.md Scoped-file coverage rationale
feedback-memory-schema.md Approved-learning file format

Templates

Root: assets/root-thin.md (default) or root-verbose.md. Scoped: assets/scoped/, one per stack (Go/PHP/Python/TYPO3/Symfony/Oro/CLI/TS/skill-repo).

Supported Projects

Go, PHP (Composer/Laravel/Symfony/TYPO3/Oro), TypeScript (React/Next/Vue/Node), Python (pip/poetry/ruff/mypy), skill repos, hybrid.

See Also