SLIDESHOT [docs] Markdown
For agents: read the markdown version of these docs →

ChatGPT setup

ChatGPT does not run shell commands or arbitrary MCP servers directly, so the Slideshot setup is slightly different from terminal agents. The most reliable paths are using the Slideshot skill inside a Project or Custom GPT, or asking ChatGPT to draft the HTTP requests and CLI commands for you to run locally.

Copy this into ChatGPT

Paste this instruction into a ChatGPT conversation, Project, or Custom GPT.

You are helping me use Slideshot from ChatGPT.

Load the Slideshot skill (https://github.com/slideshot/skills) into your context
so you understand how Slideshot recordings work.

When I describe a workflow I want to record:

1. Plan the run options I should pass: target_url, goal, and any video,
   artifacts, or intro options that fit the audience and distribution channel.
2. Draft the exact `curl` command or `slideshot` CLI command I need to run
   locally to create the run.
3. After I paste the run_id back to you, draft the next request I need to make
   to poll status and download the demo.mp4 artifact.

Do not invent endpoints or options that are not in the skill.

1. Load the skill into ChatGPT

The skills installer cannot reach ChatGPT directly, so the skill has to be pasted in by hand. Two options:

  • Project / Custom GPT instructions. Paste the contents of slideshot/SKILL.md from github.com/slideshot/skills into the Project or Custom GPT system instructions.
  • One-off conversation. Attach the skill markdown file to the conversation and tell ChatGPT to treat it as the source of truth for Slideshot.

2. Decide where Slideshot calls will happen

ChatGPT cannot create runs by itself. You have two practical options:

  • Run commands locally. ChatGPT drafts curl or slideshot CLI commands and you run them in your terminal.
  • Use the API from your own backend. Wire your service to https://api.slideshot.ai with an API key. ChatGPT helps you write the integration.

For the CLI path:

export SLIDESHOT_API_KEY="sk_live_xxxx"

npx slideshot runs create https://app.example.com \
  --goal "Open the analytics dashboard"

For raw HTTP:

export SLIDESHOT_API_KEY="sk_live_xxxx"

curl -X POST https://api.slideshot.ai/v1/agent/runs \
  -H "x-api-key: $SLIDESHOT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "target_url": "https://app.example.com", "goal": "..." }'

Test the setup

Ask ChatGPT to draft the exact command for a small recording:

Draft a slideshot runs create command for https://docs.slideshot.ai that captures the docs landing page and exports a GIF.

Run the command locally, paste the run_id back into the conversation, and let ChatGPT draft the follow-up commands to poll the run and download demo.mp4.

Troubleshooting

  • ChatGPT invents endpoints or option names. Reload the skill into the conversation. It is the source of truth for the API surface.
  • Commands fail with 401. Make sure SLIDESHOT_API_KEY is exported in the shell you are running the commands in.
  • Recording stops at a login page. Save credentials for that domain. See Login to any web app.
  • Install skills: how to load the Slideshot skill into a ChatGPT Project or Custom GPT.
  • API key: get an API key to use in the suggested commands.
  • Describe the demo flow: write prompts that produce sharper recordings.