macOS Automator

Automate macOS tasks by composing and executing Automator workflows through automator CLI and AppleScript control.

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 automator and osascript available.
• 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.

text

~/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, use automator CLI first.
• For composing or inspecting workflow internals, use Automator AppleScript commands from workflow-authoring.md.
• Only use shortcuts fallback 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 -i or -D name=value only 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.md before 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 <slug> 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