CLI: `plannotator review --help` launches the UI instead of printing help

[rrei]

Environment:

• macOS (Darwin 25.5.0), Apple Silicon
• plannotator v0.21.1, single-file binary at ~/.local/bin/plannotator
• Driven from Claude Code

Symptom: Top-level plannotator --help prints usage correctly, but the subcommands have no per-subcommand --help handler. Running plannotator review --help does not print help and does not exit - the --help token is silently ignored and the review UI launches a blocking server (opening a browser tab). The same happens when probing other subcommands such as annotate --help. This makes it easy to accidentally spawn a duplicate instance: any tool or user that probes plannotator <subcommand> --help to discover flags ends up starting a real session instead.

Expected: plannotator <subcommand> --help should print that subcommand's usage and exit 0 (or at least error on the unrecognised flag), rather than launching the UI.

Repro:

1. plannotator review --help
2. Observe: no help text, the process blocks, and the review server/UI starts.

[rrei]

Important detail i forgot to mention above: this has been happening frequently in claude-code when claude drives the cli himself.

Claude uses --help first to explore/confirm the CLI and then issues the actual command (e.g. plannotator review), which ends up opening two browser tabs instead of one. The main problem is that, when I close the extra tab, claude receives its output too, and interprets it as "no feedback -> good to proceed", or, if the session has already advanced, claude sees that and gets confused, with unpredictable consequences and a lot of wasteful token usage. In either case, the output injected by the duplicate tab causes undesired disruption of the cc session.

TL;DR

• running commands w/ --help opens duplicate browser tabs
• closing duplicate tab injects invalid signal into cc session
• costly, unintended, and unpredictable consequences