How to use the Automate This agent skill (automate-this)

What is an agent skill, and what does Automate This do?

An agent skill is a folder of instructions that an AI agent can load when a request matches the skill's purpose. The automate-this skill is defined in a single file, SKILL.md, and it is licensed under MIT.

When you hand the agent a screen recording of a repetitive task, the skill extracts still frames from the video and transcribes any spoken narration. It then reconstructs the steps you took and asks you to confirm them before proposing anything. The skill uses tools already on your computer where possible.

• Extracts frames from the video, one every 2 seconds by default
• Transcribes narration with Whisper, if you have it installed and the recording has audio
• Reconstructs the workflow and asks you to confirm it
• Proposes automation at up to three tiers: a quick win, a script, and a full automation

How do you install the Automate This skill?

Download the ZIP for the skill from the directory page and unzip it. You can install it for yourself or for a single project.

For your user account, unzip it into ~/.claude/skills/. For one project, put it in the project's .claude/skills/ folder. In the Claude apps, you can also upload the skill under Settings, then Capabilities, then Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

What do you need installed before you start?

ffmpeg is required. Without it, the skill cannot extract frames from the recording. ffprobe is bundled with ffmpeg and is used to check whether the video has an audio track.

Whisper is optional. You only need it if the recording has narration. You can install openai-whisper with pip or whisper-cpp with Homebrew. If you decline Whisper, the skill proceeds with visual analysis only.

The skill checks for these tools before analyzing a recording and only reports problems. It also checks for Python, Node, Homebrew and several common tools, so it can keep its proposals to what you already have.

• ffmpeg (required, includes ffprobe)
• Whisper via pip install openai-whisper or brew install whisper-cpp (optional, for narration)
• Python, Node, or bash, depending on the script proposed
• Optional tools the skill checks for: curl, jq, playwright, selenium, osascript, automator, crontab

How do you trigger the skill and record the task?

Record your screen while you do the task once, at the pace you normally work. Save the video file, typically on your Desktop. Then ask the agent in plain language, giving it the file path.

The skill is meant to be triggered by natural requests that describe a recording and a repetitive task. Name the file and say what you want, for example that you want the task automated or that you want to know what the steps are.

If you narrate while you work, say what you are doing and why. The skill uses the narration to find pain points and decision points.

• Name the video file path, such as a file on your Desktop
• Say you want the repetitive task automated or scripted
• Narrate the parts you find slow or tedious, if you can

What happens during an analysis?

The skill first extracts frames with ffmpeg and stores them in a per-run temporary directory with permissions set to 0700, so only your user can read them. For recordings longer than 5 minutes, it samples one frame every 4 seconds and tells you it is doing so.

It then reconstructs the process as a numbered list of steps covering the apps used, the order of actions, the data that moved between steps, and any repetition. You must confirm this list before the skill proposes automation. A wrong reconstruction produces useless automation.

• Frames are sampled every 2 seconds (every 4 seconds for recordings over 5 minutes)
• The reconstruction asks you to confirm or correct each step
• Automation is proposed only after you confirm

What automation does the skill propose?

The skill proposes up to three tiers, and not every process needs all three. Each proposal names what it automates, what stays manual, the estimated time saved, the prerequisites, the code, how to test it, and how to undo it.

Tier 1 is a quick win under 5 minutes to set up, such as a shell alias, one-liner, keyboard shortcut or AppleScript snippet. Tier 2 is a standalone script under 30 minutes to set up. Tier 3 adds scheduled execution with cron, launchd or GitHub Actions, logging, error notifications and any integration setup, under 2 hours.

• Tier 1: quick win, under 5 minutes
• Tier 2: standalone script, under 30 minutes
• Tier 3: scheduled full automation with logging, under 2 hours
• For browser work, the skill prefers a public API first, then curl or wget, then Playwright or Selenium

What should you check for safety?

Frames and audio from the recording are passed into the agent's context. That means anything visible on screen or said aloud can reach the agent, and it may be kept in its logs. Do not record sensitive logins, passwords or private data, or blur or trim those parts before you share the file.

The generated scripts can run commands and the skill is marked as running scripts. Read every script before you run it, start with the dry run the skill suggests, and keep the undo steps. The skill does not use the network itself, but the scripts it writes may call services, so review those calls too.

• Do not record logins, passwords, or private data
• Frames and audio enter the agent's context and may be stored in its logs
• Review each generated script before running it
• Run the suggested dry run first and keep the undo steps