/install automator
Setup
On first use, follow setup.md to capture activation behavior and safety preferences.
Setup is read-only. Any local file write requires explicit user confirmation.
When to Use
User needs to automate macOS tasks with Automator workflows instead of manual UI steps. Agent handles workflow execution, workflow composition, and repeatable runbooks using official Automator interfaces.
Requirements
- macOS with
automatorandosascriptavailable. - Automator app installed at
/System/Applications/Automator.app. - Explicit user confirmation before destructive or bulk operations.
Architecture
Memory lives in ~/automator/. See memory-template.md for structure.
~/automator/
├── memory.md # Activation rules and safety defaults
├── workflows.md # Known workflow paths and run arguments
├── action-catalog.md # Verified action names and categories
└── incidents.md # Failures and proven fixes
Quick Reference
Use these files when the task needs deeper detail.
| Topic | File |
|---|---|
| Setup behavior and activation | setup.md |
| Memory structure | memory-template.md |
| Execution path matrix | interface-matrix.md |
| Workflow authoring patterns | workflow-authoring.md |
| Write safety gates | execution-guardrails.md |
| Debug and recovery | troubleshooting.md |
Data Storage
All local skill data stays in ~/automator/.
Before creating or changing local files, state the write scope and ask for confirmation.
External Endpoints
| Endpoint | Data Sent | Purpose |
|---|---|---|
| None | None | This skill uses local macOS interfaces only |
No other data is sent externally.
Core Rules
1. Pick Interface by Intent, Not Convenience
- For running an existing
.workflow, useautomatorCLI first. - For composing or inspecting workflow internals, use Automator AppleScript commands from
workflow-authoring.md. - Only use
shortcutsfallback if user explicitly asks for Shortcuts conversion.
2. Validate Workflow Identity Before Execution
- Require absolute workflow path and verify it exists.
- Confirm type (
.workflow) and target operation (read, write, destructive). - If the workflow path is ambiguous, stop and ask one clarifying question.
3. Enforce Read-Before-Write for Workflow Changes
- Before editing, inspect current action list and settings.
- Apply one mutation at a time and re-read state after each mutation.
- Never batch-edit unknown actions in a single pass.
4. Parameterize Inputs with Explicit Boundaries
- Use
automator -ior-D name=valueonly with validated inputs. - Reject unbounded stdin streams for write workflows.
- Echo resolved parameters before run so the user can verify intent.
5. Require Two-Step Confirmation for Destructive Runs
- Use
execution-guardrails.mdbefore delete, reset, or mass-change paths. - Ask for explicit confirmation that includes target and scope.
- If confirmation is missing, do not run the workflow.
6. Keep Runs Observable and Reproducible
- Prefer verbose mode (
-v) for first execution or after failure. - Record command, input source, and result in concise run notes.
- Return actionable output, not only "completed" status.
7. Recover with Concrete Next Actions
- On failure, classify as path, permission, action mismatch, or runtime data error.
- Provide the next command to run, not generic retry advice.
- Persist only reusable fix patterns into local memory.
Automator Traps
- Running relative paths from unknown working directories -> workflow not found.
- Guessing action names without dictionary inspection -> compile succeeds, runtime fails.
- Feeding multiline stdin into write workflows without boundaries -> unintended bulk edits.
- Mixing Automator and Shortcuts assumptions in one run -> incompatible action model.
- Treating permission prompts as transient errors -> repeated blocked execution.
Security & Privacy
Data that stays local:
- Workflow paths, verified action names, and run diagnostics in
~/automator/. - Command output required to complete the requested automation task.
Data that leaves your machine:
- None by default.
This skill does NOT:
- Access credentials outside the workflow request scope.
- Send workflow content to third-party services.
- Execute destructive automation without explicit confirmation.
Related Skills
Install with clawhub install \x3Cslug> if user confirms:
applescript- Script app automation with robust quoting and pre-read checks.automate- Design reliable multi-step automation workflows.macos- Use macOS command-line and system operation patterns.workflow- Structure repeatable workflows and handoff checkpoints.
Feedback
- If useful:
clawhub star automator - Stay updated:
clawhub sync
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install automator - 安装完成后,直接呼叫该 Skill 的名称或使用
/automator触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
macOS Automator 是什么?
Automate macOS tasks by composing and executing Automator workflows through automator CLI and AppleScript control. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 508 次。
如何安装 macOS Automator?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install automator」即可一键安装,无需额外配置。
macOS Automator 是免费的吗?
是的,macOS Automator 完全免费(开源免费),可自由下载、安装和使用。
macOS Automator 支持哪些平台?
macOS Automator 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(darwin)。
谁开发了 macOS Automator?
由 Iván(@ivangdavila)开发并维护,当前版本 v1.0.0。