Automatic Skill — 每日 Skill 自动工厂

> 每天凌晨自动调研 → 设计 → 制作 → 审核 → 自测 → 发布 → 复查，全程无需人工介入（发布阶段使用你自己的 GitHub/clawHub 凭证）

⚠️ Credentials

Stages 1–7 use zero credentials. Only Stage 8 (Upload) and beyond touch your accounts.

Stage	Action	Credential	What gets written
-------	--------	-----------	-------------------
1–7	Research, Design, SEO, Create, Review, Self-Run, Self-Check	None	Local files only
8a	`gh repo create / --public`	`GITHUB_TOKEN`	New public repo on your GitHub account
8b	`git push` to `GITHUB_REPO`	`GITHUB_TOKEN`	New commits to the monorepo you set
8c	`clawhub publish`	`CLAWHUB_TOKEN`	Skill listed publicly under your clawHub account
9–10	Verify, Final Review	`GITHUB_TOKEN` (read-only `gh api`)	Nothing — read-only checks
Cron	Daily auto-run (`push-toggle on`)	None	Registers a local cron job via `openclaw cron`

Does NOT:

Read, transmit, or log credentials
Access repos or accounts other than the ones you configure
Execute network operations from scripts — all network actions go through gh CLI and clawhub CLI under your control
Run Stage 8 in --dry-run mode (safe for testing entire pipeline without publishing)

Minimum token scopes:

GITHUB_TOKEN: fine-grained PAT → single repo → Contents: Read & write only
CLAWHUB_TOKEN: keep in shell env (~/.zshrc), never commit to any file

Purpose & Capability

automatic-skill 是一个元技能（meta-skill），它的能力是制造其他 skill。

架构说明（重要）： 每个阶段脚本是"prompt generator" — 脚本本身不执行任何业务逻辑，只输出结构化 prompt。真正的执行者是 agent（Claude）。网络操作（GitHub push、clawHub publish）均由 agent 调用 gh CLI 和 clawhub CLI 完成，凭证完全在你的机器上。

核心能力：

能力	说明
------	------
全流水线生成	从"想法"到"上线"：Research → Design → SEO → Create → Review → Self-Run → Self-Check → Upload → Verify → Final Review
每日自动选题	凌晨 02:00 cron 自动触发，选最高价值 idea，跑完整流程
手动触发	指定 `--idea` 手动选题，或单独跑某一阶段调试
dry-run 模式	`--dry-run`：执行到 Stage 7 自检后停止，不发起任何网络操作
已有 skill 迭代	对现有 skill 单独跑 SEO / self-check / upload 阶段进行升级

能力边界（不做的事）：

不直接写 skill 源码（源码由 agent 根据 create 阶段 prompt 生成）
不并发生成多个 skill（每次运行产出一个）
不访问或修改除 GITHUB_REPO 和 SKILL_OUTPUT_DIR 以外的任何路径或账号

Instruction Scope

在 scope 内（会处理）：

"帮我自动生成一个 skill" / "skill 流水线跑了没" / "今天出了什么新 skill"
手动触发全流水线或单阶段重跑（如"重跑 SEO 阶段"）
查看流水线状态和历史日志（status.js）
开关每日定时任务（push-toggle on/off）
用 automatic-skill 迭代已有 skill（对现有 skill 升级 / SEO 优化 / 自检 / 上传）

不在 scope 内（不处理）：

直接编写 skill 源码（源码由 agent 执行 create 阶段 prompt 生成，非 automatic-skill 职责）
管理 GitHub 仓库权限、clawHub 账户设置、或任何账户级操作
生成非 openclaw skill 格式的内容（普通项目、应用代码等）
在未设置 GITHUB_TOKEN / GITHUB_REPO / CLAWHUB_TOKEN 的环境中执行 Stage 8（会提示缺少凭证，不会静默失败）

凭证缺失时的行为：

Stages 1–7：无需凭证，正常运行
Stage 8 upload：缺少任意必需凭证 → 打印具体缺失变量名 + 配置指引，停止执行（不静默失败）
--dry-run：完整跳过 Stage 8，可安全在无凭证环境测试前七个阶段

Persistence & Privilege

持久化写入的内容：

路径	内容	触发条件
------	------	---------
`SKILL_OUTPUT_DIR//`	生成的 skill 文件	Stage 4 Create
`data/current-pipeline.json`	当前流水线运行状态（临时）	每次流水线运行
`data/pipeline-log.json`	历次运行历史（追加写入）	Stage 10 Final Review
`~/.openclaw/crontab`（或系统 cron）	每日定时任务条目	`push-toggle on` 时写入，`off` 时删除

不写入的路径：

不修改 shell 配置文件（~/.zshrc 等）
不写入 GITHUB_TOKEN / CLAWHUB_TOKEN 等凭证到任何文件
不在 SKILL_OUTPUT_DIR 之外创建文件（除 data/ 目录）

权限级别：

运行时以当前用户身份执行，不需要 sudo 或提权
gh CLI 和 clawhub CLI 的权限上限由你配置的 token 范围决定
如需撤销所有权限：push-toggle off（删除 cron）+ 在 GitHub/clawHub 吊销对应 token

Install Mechanism

标准安装（推荐）

从 clawHub 安装到 openclaw workspace：

clawhub install automatic-skill
# 安装路径：~/.openclaw/workspace/skills/automatic-skill/

手动安装

从本地源码目录复制：

cp -r /path/to/automatic-skill ~/.openclaw/workspace/skills/automatic-skill/

验证安装

ls ~/.openclaw/workspace/skills/automatic-skill/scripts/
node ~/.openclaw/workspace/skills/automatic-skill/scripts/status.js
# 应输出：No active pipeline run. / pipeline-log.json 状态

安装后配置

在 shell 环境中设置以下变量（写入 ~/.zshrc 或 ~/.bashrc）：

export GITHUB_TOKEN=<your-github-token>          # repo write scope
export GITHUB_REPO=<owner/repo>                  # e.g. Cosmofang/openclaw-skills
export CLAWHUB_TOKEN=<your-clawhub-token>
export CLAWHUB_OWNER_ID=<your-owner-id>          # optional
export SKILL_OUTPUT_DIR=~/.openclaw/workspace/skills  # optional, this is the default

然后验证认证：

gh auth status        # 应显示 ✓ Logged in
clawhub whoami        # 应返回你的用户名

启用每日自动运行

node scripts/push-toggle.js on
# 注册 cron：每天 02:00 自动跑 daily-pipeline.js

何时使用

用户说"帮我自动生成一个 skill"/"今天有新 skill 吗"/"skill 流水线跑了没"
用户想查看今日生成的 skill 是什么
用户想手动触发某一阶段重跑
用户想查看生成历史或当前流水线状态
用户想用 automatic-skill 迭代/升级/上传某个现有 skill

🔄 10-Stage Pipeline

#	Stage	Script	Description
---	-------	--------	-------------
1	Research	`research.js`	Scan trends, identify skill gaps, output Top-3 ideas
2	Design	`design.js`	Produce full architecture: file tree, script specs, data schema
3	SEO	`seo.js`	Optimize display name, tagline, description, 30+ keywords
4	Create	`create.js`	Generate all files using design + SEO output
5	Review	`review.js`	Quality checklist: structure, scripts, content, security
6	Self-Run	`self-run.js`	Execute every script, verify zero errors
7	Self-Check	`self-check.js`	Validate required fields, file tree, script signatures
7.5	Safety Check	`scan-check.js`	Pre-upload credential-safety gate; blocks upload if issues found
8	Upload	`upload.js`	Create standalone repo + push monorepo; clawhub publish
9	Verify	`verify-upload.js`	Confirm live on GitHub and clawHub
10	Final Review	`final-review.js`	Full report, write to pipeline-log.json

📦 Publishing Convention

Skill Standard Sections

Every skill generated or uploaded by automatic-skill must include all five of the following sections in its SKILL.md, in this order:

#	Section	Required content
---	---------	-----------------
1	`## Purpose & Capability`	Core concept + capability table + explicit "Does NOT" boundary
2	`## Instruction Scope`	In-scope examples + out-of-scope list + behavior on missing credentials
3	`## Credentials`	Action/Credential/Scope table (or "no credentials required") — no hardcoded tokens
4	`## Persistence & Privilege`	Paths-written table + "Does NOT write" list + uninstall instructions
5	`## Install Mechanism`	`clawhub install` command + verification step + env var examples

Stage 7 (self-check.js) validates all five sections before allowing upload. Missing or empty section = NEEDS_FIX.

GitHub Distribution

Every skill uploaded by automatic-skill must have both:

Destination	Purpose	Command
-------------	---------	---------
`/` (standalone repo)	Users can search and find the skill directly on GitHub	`gh repo create / --public` → push skill files
`GITHUB_REPO/openclaw/agents/skills//` (monorepo)	Registry index — all skills listed in one place	`git add openclaw/agents/skills//` → push

Stage 8 always runs standalone repo creation first, then monorepo push.

If the standalone repo already exists, skip creation and force-push the latest files.

🌐 Language Policy

SKILL.md content: English (default)
Conversation with user: match user's language — Chinese if user writes Chinese
JSON logs & reports: English only

🛠️ Usage

# Full pipeline (recommended)
node scripts/pipeline.js                   # auto-select topic, run all stages
node scripts/pipeline.js --idea "daily-poem" # specify topic
node scripts/pipeline.js --dry-run         # run through self-check only, no upload

# Per-stage debug
node scripts/research.js                   # output research prompt
node scripts/design.js "daily-poem"        # output design prompt
node scripts/seo.js --from-pipeline        # output SEO optimization prompt
node scripts/create.js --from-pipeline     # output create prompt
node scripts/review.js /path/skill-dir     # output review prompt
node scripts/self-run.js /path/skill-dir   # output self-run prompt
node scripts/self-check.js /path/skill-dir # output self-check prompt
node scripts/upload.js /path/skill-dir     # output upload prompt (direct push)
node scripts/upload.js --pr /path/skill-dir  # output upload prompt (PR workflow)
node scripts/verify-upload.js skill-name   # output verify prompt (gh api checks)
node scripts/final-review.js skill-name    # output final review prompt

# Status & history
node scripts/status.js                     # current pipeline status
node scripts/status.js --history [N]       # last N runs (default 10)
node scripts/status.js --clear             # clear current pipeline state

# Cron toggle
node scripts/push-toggle.js on             # enable daily 02:00 auto-run
node scripts/push-toggle.js off            # disable
node scripts/push-toggle.js status         # show status

⏰ Cron Setup

openclaw cron add "0 2 * * *" "cd ~/.openclaw/workspace/skills/automatic-skill && node scripts/daily-pipeline.js"
openclaw cron list
openclaw cron delete <job-id>

📁 File Structure

data/
  pipeline-log.json       # append-only history of all pipeline runs
  current-pipeline.json   # transient state during active run
scripts/
  research.js             # stage 1
  design.js               # stage 2
  seo.js                  # stage 3 — SEO optimization
  create.js               # stage 4
  review.js               # stage 5
  self-run.js             # stage 6
  self-check.js           # stage 7
  upload.js               # stage 8
  verify-upload.js        # stage 9
  final-review.js         # stage 10
  pipeline.js             # orchestrator
  daily-pipeline.js       # cron entry point
  status.js               # status query
  push-toggle.js          # cron toggle

⚠️ Notes

Required env vars: GITHUB_TOKEN, GITHUB_REPO, CLAWHUB_TOKEN
Optional: CLAWHUB_OWNER_ID (your clawHub owner ID), SKILL_OUTPUT_DIR (default: ~/.openclaw/workspace/skills)
GitHub operations use the gh CLI (inspired by steipete/github): gh auth status, gh repo view, gh api, gh pr create. Install with brew install gh and run gh auth login before first use.
Stage 8 supports --pr flag for a PR-based GitHub workflow instead of direct push to main.
Stage 9 verifies GitHub state live via gh api repos/{repo}/contents/{path} — no git fetch needed.
Every skill gets a standalone GitHub repo (/) in addition to the monorepo entry — this lets users search and discover skills directly on GitHub
On any stage failure the pipeline stops and logs the error — no partial overwrites
--dry-run stops after self-check, no network operations
Full run history in data/pipeline-log.json
All scripts are prompt generators only — no outbound network requests are made by the scripts themselves

🧪 Test Run Log (2026-04-04)

First full dry-run results:

Stage	Status	Notes
-------	--------	-------
1 Research	✅	Scanned 17 existing skills, identified 5 gaps, selected daily-poem
2 Design	✅	Fix: `--from-pipeline` reads from `pipeline.research.selected` not top-level
3 SEO	✅	Added in v1.1.0 — produces displayName, tagline, 30+ keywords
4 Create	✅	Generated 9 files including 4 scripts
5 Review	✅	Score 97/100, 1 warning (empty data array, expected)
6 Self-Run	✅	6 script calls, all passed
7 Self-Check	✅	22/22 checks passed, score 100
8 Upload	✅	dry-run skipped — requires GITHUB_TOKEN / GITHUB_REPO / CLAWHUB_TOKEN
9 Verify	✅	dry-run skipped
10 Final Review	✅	Archived to pipeline-log.json

Environment variables:

export GITHUB_TOKEN=<your-token>
export GITHUB_REPO=<owner/repo>
export CLAWHUB_TOKEN=<your-clawhub-token>
export CLAWHUB_OWNER_ID=<your-owner-id>   # optional
export SKILL_OUTPUT_DIR=~/.openclaw/workspace/skills  # optional

Version: 1.4.1 · Created: 2026-04-04 · Updated: 2026-04-09

Automatic Skill

概述