Writ

Writ connects to Claude, Codex, Cursor, or OpenCode through the agent’s native hook. You keep using the same agent. The first connection needs no writ file, model key, or MCP configuration.

you → your agent → native hook → Writ decision + record
                        ↓
                   agent runs tool

Choose one level

Basic: record calls

Run writ connect <agent>. Writ records the tool calls that the agent reports. It blocks nothing. Connect and review.

Normal: add a limit

Allow read and search tools, or name exact tools. Writ stops other calls for approval. Add a tool limit.

Advanced: apply a writ

Apply a writ to one selected session of your existing agent. Use the managed adapter only when the task needs its model loop and tools. Choose the workflow.

Start with one command

Run the command for your agent once:

writ connect claude
writ connect codex
writ connect cursor
writ connect opencode

Start or restart the agent. Use it as usual. Review the run after a tool call:

writ review

The default connection applies to your user account. Run writ connect <agent> --project when only the current project needs the connection. Writ keeps project records under ~/.writ/projects/. It does not create a .writ directory in the project.

The connection status states which built-in, local MCP, remote MCP, and hosted tools the selected agent reports. Tool results come from the agent. Writ does not independently prove their effects.