agent-skills
AGENTS.md1 starsChanged 17 days agoMITView on GitHub
个人 AI coding agent skills 集合:每个 skill 位于 skills/ 下一个目录,通过 junction/软链分发到各 agent 的 skills 目录(目标 = 内置主流清单 ⊕ agents.local.conf,目录存在才生效,清单见 docs/CONVENTIONS.md §6)。 纯提示词 skill(deep-research、theme-commit、zero-coding、session-handoff)无运行时,直接读 SKILL.md 并按 docs/TESTS.md 人工回归。 视频单测 163 条,真实托管 API 用例需 VOLC_BIGMODEL_API_KEY / PADDLEOCR_ACCESS_TOKEN,未设置自动跳过。windows-disk-cleanup 单测 42 条,真实扫描/回收站用例需在 Windows 人工执行。md2pdf 改动按 skills/md2pdf/docs/TESTS.md 回归,核心质量门为 S-3 canary 双版式。 架构与依赖细节写在各 skill 的 README.md / docs/SPEC.md(md2pdf 为 docs/SPEC.md + docs/DECISIONS.md),本文不重复。 skills-lint.sh exit 0 且 0 error;受影响 skill 的测试集已回归(代码型另跑自动化测试);链接变更后 doctor 无 broken;版本在 SKILL.md / pyproject.toml / plugin.json / marketplace.json / 台账间一致。
- Reads credentials
In this repository9 other files it gives its agents.
- ddgs-web-accessSkill
- deep-researchSkill
- md2pdfSkill
- session-handoffSkill
- theme-commitSkill
- video-insightSkill
- windows-disk-cleanupSkill
- zero-codingSkill
- __NAME__Skill
The file
Files
AGENTS.md — skills
个人 AI coding agent skills 集合:每个 skill 位于 skills/ 下一个目录,通过 junction/软链分发到各 agent 的 skills 目录(目标 = 内置主流清单 ⊕ agents.local.conf,目录存在才生效,清单见 docs/CONVENTIONS.md §6)。
改前必读
- docs/CONVENTIONS.md — 目录、frontmatter、版本、分发、脚本规则、提交信息。
- spec/agent-skills-spec.md — 官方规范指针与本仓库加严项。
- 目标 skill 的
README.md与docs/TESTS.md;新增 skill 先用template/或scripts/skills-new.sh。
怎么跑起来 / 验证
纯提示词 skill(deep-research、theme-commit、zero-coding、session-handoff)无运行时,直接读 SKILL.md 并按 docs/TESTS.md 人工回归。
# 仓库级校验(纯 bash,零依赖;退出码 0/1/2)
bash scripts/skills-lint.sh # 结构/元数据:全绿 exit 0
bash scripts/skills-doctor.sh # 链接健康矩阵
bash scripts/skills-link.sh --all # 分发(幂等)
# ddgs-web-access(uv >= 0.5;调用间隔 >=2 秒)
bash skills/ddgs-web-access/bin/ddgs-web-search "smoke test" -m 1
bash skills/ddgs-web-access/bin/ddgs-web-fetch "https://example.com"
# md2pdf(uv >= 0.5 + 系统 Edge;首次 fetch 约 15MB)
bash skills/md2pdf/bin/md2pdf --help
bash skills/md2pdf/bin/md2pdf fetch
# video-insight(纯标准库;外部依赖 ffmpeg/ffprobe)
python -m unittest discover -s skills/video-insight/tests
# windows-disk-cleanup(纯标准库 + PowerShell 辅助;uv 可选)
python -m unittest discover -s skills/windows-disk-cleanup/tests
python skills/windows-disk-cleanup/scripts/memory.py check --json
bash skills/windows-disk-cleanup/bin/windows-disk-cleanup doctor --root . --json视频单测 163 条,真实托管 API 用例需 VOLC_BIGMODEL_API_KEY / PADDLEOCR_ACCESS_TOKEN,未设置自动跳过。windows-disk-cleanup 单测 42 条,真实扫描/回收站用例需在 Windows 人工执行。md2pdf 改动按 skills/md2pdf/docs/TESTS.md 回归,核心质量门为 S-3 canary 双版式。
skill 矩阵
| Skill | 形态 | 入口 | 验证 |
| ddgs-web-access | Python CLI(uv) | skills/ddgs-web-access/SKILL.md | 两个 bin 冒烟 + docs/TESTS.md |
| deep-research | 纯提示词 | skills/deep-research/SKILL.md | docs/TESTS.md(人工会话) |
| theme-commit | 纯提示词 | skills/theme-commit/SKILL.md | docs/TESTS.md(人工会话) |
| zero-coding | 纯提示词 | skills/zero-coding/SKILL.md | docs/TESTS.md(人工会话) |
| session-handoff | 纯提示词 | skills/session-handoff/SKILL.md | docs/TESTS.md(人工会话) |
| video-insight | Python 脚本(标准库) | skills/video-insight/SKILL.md | unittest + docs/TESTS.md |
| md2pdf | Python CLI(uv) | skills/md2pdf/SKILL.md | --help + canary S-3 + docs/TESTS.md |
| windows-disk-cleanup | Python CLI(标准库 + PowerShell) | skills/windows-disk-cleanup/SKILL.md | unittest + memory check + docs/TESTS.md |
架构与依赖细节写在各 skill 的 README.md / docs/SPEC.md(md2pdf 为 docs/SPEC.md + docs/DECISIONS.md),本文不重复。
目录与约定
- 一个 skill 一个目录,
SKILL.md为入口;frontmatter 需含name(== 目录名)、三段式description、metadata.version(semver,版本唯一权威)与metadata.category(建议)。 - 每个 skill 必须有
README.md与.claude-plugin/plugin.json,并在.claude-plugin/marketplace.json有对应条目;应带docs/TESTS.md(缺失 lint warn)。 - 详细条件分支、模板、长清单放
references/(距SKILL.md一层);SKILL.md目标 ≤500 行 / ≤16KB,只留核心流程与"何时读哪个文件"。 - README 的
## Skills表是唯一公开目录,列固定Skill | 说明,不展示类型与版本号。 - 维护者个人的台账、分发状态与下一步计划记录在
SKILLS.local.md(gitignore*.local.md,不入库)。 - 分发与卸载统一用
bash scripts/skills-link.sh/skills-unlink.sh;链接变更后跑 doctor 并回填台账。内置目标清单的权威实现是scripts/lib.sh的builtin_agent_targets()。 - 运行产物(
.venv/、.cache/、__pycache__/、*.egg-info/)已 gitignore,lint 检查不得入库;.zcode/同样排除。 - 不改写 Git 历史、不推送;提交信息遵循规范 §10(
type(scope): 摘要+ 列表式 body)。
完成定义
skills-lint.sh exit 0 且 0 error;受影响 skill 的测试集已回归(代码型另跑自动化测试);链接变更后 doctor 无 broken;版本在 SKILL.md / pyproject.toml / plugin.json / marketplace.json / 台账间一致。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

