Module app 

CLI entry orchestrator. Mirrors the v1-relevant slice of the TS packages/coding-agent/src/main.ts — the main(args) function that:

1. Parses argv (crate::args::parse_args).
2. Handles --help/--version + parse errors + startup warnings.
3. Reads piped stdin (non-TTY ⇒ treat as the initial prompt text — TS readPipedStdin).
4. Expands @file attachments into an initial-message text block (TS processFileArguments + [build_initial_message] — the port of TS buildInitialMessage).
5. Resolves the provider + model + thinking level (crate::provider::resolve).
6. Builds the harness (crate::session::build).
7. Resolves the effective run mode (crate::args::resolve_mode) and dispatches to crate::modes (print/json/interactive), mapping the outcome to an exit code.

v1 scope cuts vs TS main.ts (in docs/m6-cli-open-questions.md)

The TS main is enormous: auth-command routing, package-manager commands, HTTP proxy config, project-trust prompts, first-time setup, migrations, settings managers, theme init, extension/resource discovery. None of that is ported — v1 is a straight parse → resolve → build → run pipeline. The @file expansion ports only the text-file branch (images are detected but not attached to the prompt — the harness prompt_text accepts images, but v1 does not yet wire an image processor; binary/non-UTF-8 files error).

Constants

EXIT_RUNTIME

The exit code for a runtime failure (model-resolution, harness-build, or run failure). Mirrors TS process.exitCode set from runPrintMode.

EXIT_USAGE

The exit code for a usage/parse error. (TS main.ts uses process.exit(1) for most error paths; v1 distinguishes usage errors with the conventional 2 so scripts can tell “bad invocation” from “run failed”.)

Functions

run

The v1 CLI entry point. Mirrors TS export async function main(args).