Continue CLI (cn) Cheatsheet - Terminal Open-source Coding Agent

Use -p for a single run that prints and exits, with CONTINUE_API_KEY for headless authentication so it does not depend on a local login. Before a large change, run once with --readonly to get a plan and confirm the approach before letting it write. Add --auto only when you genuinely want unattended approval of every tool call.

Yes. @src/auth/middleware.ts feeds exactly that file into context — precise and cheap in tokens — whereas a full-project scan is slow, pulls in a lot of irrelevant content, and can lead it to advise based on code that has nothing to do with the task. When cross-file understanding is needed, reference the few key entry files rather than the whole directory.

Yes. They share the same agent engine, so model selection, MCP servers and assistant configuration carry over. Habits built up in the TUI continue to work in the IDE, which keeps switching cost low. For the same reason, a configuration you have validated locally can be committed to a config file that both sides read, preventing the two from drifting apart.

cn --mcp <slug> mounts an MCP server so the agent can call external tools, and cn --agent <slug> loads a custom agent file with predefined prompts and flow. These configurations are shared with the Continue IDE plugins, so setting them up once serves both. Before CI, verify with readonly mode and only then allow unattended approval.

Use the --rule flag, for example cn --rule "use TypeScript strict mode", which injects the constraint so it stays in effect across turns instead of being repeated in every prompt. Compared with a system prompt, rules are finer-grained and suit checkable requirements such as language choice or code style.