Developers · CLI
Work where your project lives.
The spiron CLI opens an interactive terminal agent in your project. Use a one-shot prompt for a bounded task, or continue an existing session.
- Executable
- spiron
- Script output
- JSONL agent events
On this page3 sections
01 · Launch
Start with your installed build.
spiron --help spiron version --json spiron login spiron whoami spiron
These commands assume spiron is already installed and on your PATH. With no subcommand, it opens the interactive TUI in the current project. Sign-in may require completing the device flow in your browser.
Building from a repository checkout instead? With the Rust toolchain and project prerequisites available, inspect the CLI with:
cargo run -p spiron-cli --bin spiron -- --help
02 · Prompt and continue
Choose the output your caller needs.
spiron -p "Read this project and summarize how to run its tests. Do not edit files." spiron --json -p "List the top-level directories. Do not edit files." spiron --session "$SPIRON_SESSION_ID"
-p supplies a one-shot user message. --json emits agent stream events as JSON Lines; it is not one JSON result object. For the last command, set SPIRON_SESSION_ID to a real existing session ID.
A request to avoid edits guides the agent; it does not replace tool permissions. Restrict access to match the task and inspect the result yourself.
03 · Configure
Make the project boundary explicit.
| Option | Purpose |
|---|---|
| --workdir PATH | Select the project root before discovering instructions and skills. |
| --mcp-config PATH | Load JSON stdio MCP server mounts using a mcpServers object. |
| --skills-dir PATH | Add a skill directory; workspace .agents/skills is also discovered. |
| --profile PATH_OR_NAME | Select a TOML config path or a named profile in the private state directory. |
| --state-dir PATH | Override the private config, login, and session root; takes precedence over SPIRON_HOME. |
Only mount MCP servers you trust. Keep private state out of the project repository. Use spiron --help as the reference for the build you are actually running; flags in main may be newer.
Try it
Give the agent a checkable task.
Start read-only. Review the result before granting broader access.