Headless Mode

Run cn -p "your prompt" to execute a single task without an interactive session. The agent runs to completion and prints its response to stdout.

cn -p "Generate a conventional commit message for the staged changes"

Piping

Pipe data into cn to include it as context:

git diff --staged | cn -p "Write a commit message for this diff"
cat error.log | cn -p "Explain what went wrong"
curl -s https://api.example.com/health | cn -p "Is this healthy?"

Pipe the output into other tools:

cn -p "Generate a commit message" | git commit -F -
cn -p "List all TODO comments in src/" --silent > todos.txt

Output options

Flag	Effect
--silent	Strip <think> tags and excess whitespace from output
--format json	Output structured JSON

--silent is useful when piping output into another program. --format json is useful for parsing results programmatically.

Tool permissions

In headless mode, tools that would normally prompt for approval (ask permission) are automatically excluded — there’s no one to approve them. To enable write operations, explicitly allow them:

# Allow the agent to write files
cn -p "Fix the type errors in src/" --allow Write --allow Edit

# Allow everything
cn -p "Set up the project" --allow "*"

See tool permissions for the full policy system.

Resume in headless mode

Resume a previous session without interactive input:

cn -p --resume

This replays the previous session’s history, which can be useful for inspecting past results in automation.

Examples

# Git hooks
cn -p "Review staged changes for obvious bugs" --silent

# CI pipeline — allow file writes for auto-fix
cn -p "Fix all ESLint errors in src/" --allow Write --allow Edit --allow Bash

# Generate docs from code
cn -p "@src/api/ Generate OpenAPI documentation for these endpoints" --silent > api-docs.yaml

# Use a specific assistant
cn -p "Triage this error log" --config my-org/error-triage-assistant