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.

StageActionCredentialWhat gets written1–7Research, Design, SEO, Create, Review, Self-Run, Self-CheckNoneLocal files only8agh repo create <owner>/<slug> --public``GITHUB_TOKENNew public repo on your GitHub account8bgit push to GITHUB_REPO``GITHUB_TOKENNew commits to the monorepo you set8cclawhub publish``CLAWHUB_TOKENSkill listed publicly under your clawHub account9–10Verify, Final ReviewGITHUB_TOKEN (read-only gh api)Nothing — read-only checksCronDaily auto-run ( push-toggle on)NoneRegisters 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/<slug>/生成的 skill 文件Stage 4 Createdata/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：

bash

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：

bash

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

验证安装#

bash

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）：

bash

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：

bash

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#

bash

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#

#StageScriptDescription1Researchresearch.jsScan trends, identify skill gaps, output Top-3 ideas2Designdesign.js <idea>Produce full architecture: file tree, script specs, data schema3SEOseo.jsOptimize display name, tagline, description, 30+ keywords4Createcreate.jsGenerate all files using design + SEO output5Reviewreview.js <skill-dir>Quality checklist: structure, scripts, content, security6Self-Runself-run.js <skill-dir>Execute every script, verify zero errors7Self-Checkself-check.js <skill-dir>Validate required fields, file tree, script signatures7.5Safety Checkscan-check.js <skill-dir>Pre-upload credential-safety gate; blocks upload if issues found8Uploadupload.js <skill-dir>Create standalone repo + push monorepo; clawhub publish9Verifyverify-upload.js <skill-name>Confirm live on GitHub and clawHub10Final Reviewfinal-review.js <skill-name>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:

#SectionRequired content1## Purpose & CapabilityCore concept + capability table + explicit "Does NOT" boundary2## Instruction ScopeIn-scope examples + out-of-scope list + behavior on missing credentials3## CredentialsAction/Credential/Scope table (or "no credentials required") — no hardcoded tokens4## Persistence & PrivilegePaths-written table + "Does NOT write" list + uninstall instructions5## 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:

DestinationPurposeCommand<owner>/<slug> (standalone repo)Users can search and find the skill directly on GitHubgh repo create <owner>/<slug> --public → push skill filesGITHUB_REPO/openclaw/agents/skills/<slug>/ (monorepo)Registry index — all skills listed in one placegit add openclaw/agents/skills/<slug>/ → 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#

bash

# 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#

bash

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#

text

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#

1. Required env vars: GITHUB_TOKEN, GITHUB_REPO, CLAWHUB_TOKEN
2. Optional: CLAWHUB_OWNER_ID (your clawHub owner ID), SKILL_OUTPUT_DIR (default: ~/.openclaw/workspace/skills)
3. 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.
4. Stage 8 supports --pr flag for a PR-based GitHub workflow instead of direct push to main.
5. Stage 9 verifies GitHub state live via gh api repos/{repo}/contents/{path} — no git fetch needed.
6. Every skill gets a standalone GitHub repo ( <owner>/<slug>) in addition to the monorepo entry — this lets users search and discover skills directly on GitHub
7. On any stage failure the pipeline stops and logs the error — no partial overwrites
8. --dry-run stops after self-check, no network operations
9. Full run history in data/pipeline-log.json
10. 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:

StageStatusNotes1 Research✅Scanned 17 existing skills, identified 5 gaps, selected daily-poem2 Design✅Fix: --from-pipeline reads from pipeline.research.selected not top-level3 SEO✅Added in v1.1.0 — produces displayName, tagline, 30+ keywords4 Create✅Generated 9 files including 4 scripts5 Review✅Score 97/100, 1 warning (empty data array, expected)6 Self-Run✅6 script calls, all passed7 Self-Check✅22/22 checks passed, score 1008 Upload✅dry-run skipped — requires GITHUB_TOKEN / GITHUB_REPO / CLAWHUB_TOKEN9 Verify✅dry-run skipped10 Final Review✅Archived to pipeline-log.json

Environment variables:

bash

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