Skip to content
offprompt

Introduction

offprompt lets a coding agent ask you for a secret and have it land where the project needs it. The agent says what it needs and why; a page opens in your browser; you type the value there. The transcript, the model's context and the provider's logs only ever see the secret's name.

Why

An agent that needs an API key usually asks for it in the chat. Whatever you paste there goes into the model's context, the transcript on disk and the provider's logs, and stays there. offprompt gives the agent another way to ask: a tool call that opens a page in your browser, from which the value goes straight into the file.

How it works

The agent asks

It calls collect_secret with every key it needs, why it needs them, and the file they go in. Then it waits: its turn stops inside the call.

A page opens

offprompt opens a page on 127.0.0.1 in your browser, or, from a cloud sandbox, gives the agent a link to it. The page says which agent is asking, what for, and which file it writes to. Each field knows what its value should look like, and links to where the key is made.

You insert the values

Each value is checked as it goes in, typed, pasted or imported from a file. When every field passes, you press the write button.

The file is written

offprompt writes the values and answers the agent with their names, the file and a fingerprint of what the file now holds. The agent reads them from the environment by name from then on.

What the agent gets back

The whole result of the call, with no value in it. It opens with the line the agent is asked to write to you:

{
  "tell_user": "Fingerprint 🍋 🥝 🌮 🧀",
  "request_id": "9f3c1e7a",
  "status": "written",
  "expires_in": 271,
  "names": ["RESEND_API_KEY"],
  "emoji": "🍋 🥝 🌮 🧀",
  "file_keys": ["DATABASE_URL", "RESEND_API_KEY"],
  "sink": { "kind": "dotenv", "path": ".env" },
  "note": "Write this to the user now, before your next tool call, on a line of its own: \"Fingerprint 🍋 🥝 🌮 🧀\". …"
}

Hosts that show the result's text, such as Codex and Pi, show one more line above the JSON: Wrote 1 value to .env · fingerprint 🍋 🥝 🌮 🧀.

The note asks the agent to tell you the fingerprint straight away, so you can check it against the four emoji on your page while it is still open, and not to open the file: reading it would put the values into the transcript, which is the one thing the page avoided.

What offprompt does, and when

offprompt acts when the agent calls it, and at no other time. It does not read your prompts, watch the agent's tool calls or rewrite what comes back from them. The agent is trusted: the problem offprompt solves is narrow, an agent needing a string only you have, in a place the project reads.

The one line it holds is at its own write. A value only goes into a file git tracks when you tick the override on the page.

Next

Suggest a change to this page