rc schema silently returns the root schema for an unknown command arg

What happens

rc commands --json emits capability IDs in colon form (apps:create, products:store:plan). rc schema expects space-separated tokens (rc schema apps create). Feeding the colon form back in silently returns the root rc schema with exit code 0 and no error:

rc schema apps create   # ✅ returns the apps-create schema
rc schema apps:create   # ⚠️ returns the ROOT schema (name: "rc"), exit 0, no error

Why it matters

An agent's natural discovery loop is: run rc commands, copy an identifier, run rc schema <that identifier>. Since rc commands gives colon IDs, the agent pastes apps:create and gets a valid-looking-but-empty root schema. It concludes the command takes no args/flags and either guesses or gives up. The failure is silent, so it's hard to notice.

Found while auditing the AI Toolkit skills: multiple agents hit this independently.

Expected

rc schema <unknown-or-colon-joined-arg> should exit non-zero with an error (e.g. "unknown command: apps:create; did you mean apps create?") rather than falling back to the root schema.

Workaround / note

• rc schema apps create (space form) works correctly.
• rc commands --schemas --json returns per-command detail for everything in one call and is the more reliable agent path.

Low severity (a doc/UX papercut, not data loss), but it defeats the primary "ask the binary" discovery flow for agents.

Start with the rc schema and rc commands --json entry points and reproduce the colon-form and space-separated cases described in the issue. Trace how command arguments are resolved, then add coverage for an unknown or colon-joined argument. Done means invalid input exits non-zero with an error instead of returning the root schema, while the space-separated form still works.