Use RecordCue with your AI agent

Install one portable Skill. Keep the CLI inside the Mac App Store app.

RecordCue works with local AI agents that can run a command and read files on this Mac. It does not require one particular agent, a RecordCue account, or a second CLI installation.

Where it works

ChatGPT desktop in a local Codex workspace, Codex CLI, and the Codex IDE extension

Supported when the Codex session runs on this Mac. Codex discovers personal Skills in ~/.agents/skills.

Claude Code

Supported when Claude Code runs on this Mac. It discovers personal Skills in ~/.claude/skills.

Other local AI agents

Supported when the agent can read an Agent Skills package, run a local shell command, and read the returned transcript or recording.

A ZIP alone cannot connect a cloud session to files and apps on your Mac. Ordinary ChatGPT web or mobile chats, Claude.ai, Claude Cowork, and other cloud sessions do not get automatic RecordCue discovery from this package alone. Use a local agent session on the Mac, or attach a transcript or recording yourself when you explicitly want the cloud service to process it.

Copy one request — no ZIP needed

1. Install and open RecordCue from the Mac App Store.
2. Copy the setup request below, or copy it from RecordCue Settings → AI tools.
3. Paste the request into the local AI agent you already use.
4. Review the destination and files it proposes, then approve the installation.

I have RecordCue installed on this Mac. Follow the setup instructions and install the RecordCue Agent Skill from the inspectable plain-text files listed at:
https://www.recordcue.app/agent/

Do not download or open the ZIP unless I explicitly ask. Before writing anything, tell me which Skill directory and files you will create. Copy the published files exactly, save the wrapper as scripts/recordcue, make it executable, and verify that it can query the latest finished recording. Use the Skill to read finished local RecordCue meetings when I ask. Prefer the local transcript unless I explicitly ask you to analyze the original audio or video. Use only your normal model processing; do not send recordings, transcripts, or results to an additional service or destination unless I explicitly ask. If you do not support Agent Skills but can run a local shell, follow the fallback instructions on that page. If you cannot read that page, or this session cannot run commands and read files on this Mac, do not claim the integration is installed; explain what is unavailable.

The agent installs from these four plain-text files. You can inspect every one before approving anything:

• SKILL.md → recordcue/SKILL.md
• recordcue-wrapper.txt → recordcue/scripts/recordcue, made executable
• cli-reference.md → recordcue/references/cli.md
• openai.yaml → recordcue/agents/openai.yaml for compatible OpenAI clients

Visiting these links only displays text. Nothing runs until your local agent creates the files and you allow it to test the wrapper.

Or install the ZIP by hand

Download the RecordCue Agent Skill ZIP, unzip it, and place the resulting recordcue folder at ~/.agents/skills/recordcue for Codex, or at ~/.claude/skills/recordcue for Claude Code. For another agent, use its documented user-level Skill directory. Restart or reload the agent if the new Skill does not appear immediately.

If your local agent can run commands but has no Skill directory, save the four plain-text files in the structure shown above somewhere stable and give the agent the path to recordcue/SKILL.md. It can follow those instructions and run recordcue/scripts/recordcue directly, although it may not discover the workflow automatically in later chats.

What happens after installation

You can ask naturally:

• “Summarize my latest meeting from its transcript.”
• “List the decisions and next actions from yesterday's calls.”
• “Analyze the original audio directly and review how I spoke.”

The Skill runs its own small wrapper. That wrapper finds the RecordCue app by its normal location or its macOS bundle identifier, then calls the read-only CLI already inside the app. You never need to find the app's executable path or install a global recordcue command.

Transcript or original recording

Transcript-based work uses the text macOS created on your Mac. It is the default when you do not specify a source. Direct analysis uses the original audio or video only when you explicitly ask for it. That request allows your current agent to use its normal media-analysis path under the permissions you already gave it; using a different third-party service still requires your approval.

The Skill runs when you ask; it does not automatically wake up when a recording ends. A scheduled or watch-folder workflow is a separate, explicit setup. Browser-only AI chats cannot read files on your Mac.

For that, see the hook that runs a script when a recording finishes and the automation recipes built on it. The full JSON contract is in the Agent CLI reference.