/ultraplan — bare-keyword auto-trigger launches a billed cloud session; opaque failures; assistant lacks usage knowledge

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

I'm a paying Claude Max customer using Claude Code (Opus, macOS). Three problems with /ultraplan wasted my paid resources:

1. Bare-keyword trigger is a cost footgun. Per the docs, including the word "ultraplan" anywhere in a normal prompt launches the billed cloud feature — so it can't even be discussed without risk of an accidental, paid cloud teleport. Expensive cloud actions should require an explicit /ultraplan invocation + confirmation, never a bare word in prose.
2. Preconditions aren't surfaced; failures are opaque. It requires a GitHub remote and an undocumented repo-size limit. The only error was "Repo is too large to teleport" — no size, no limit, no docs link, no mention of the GitHub-remote requirement. I found both by trial and error.
3. The assistant had no grounded knowledge of the feature it steered me toward (not a CLI command, not a skill, needs GitHub + a prompt arg + size cap) — it reverse-engineered this over many turns, all on my paid tokens.

What Should Happen?

1. No accidental triggering. It should run only when explicitly invoked as /ultraplan, and should ask for a confirmation before starting a billed cloud session — never fire just because the word appears somewhere in a normal prompt.
2. Preconditions checked up front, with clear errors. Before attempting to teleport, it should verify its requirements and fail with actionable messages, e.g. "A GitHub remote is required" and "Repo is 3.5 GB, which exceeds the X GB teleport limit — see <docs link>", instead of the bare "Repo is too large to teleport" with no number, no limit, and no mention of the GitHub-remote requirement.
3. The assistant should know how the feature works. Claude should have accurate built-in knowledge of /ultraplan's preconditions (GitHub remote, prompt argument, size limit) so it doesn't steer users into repeated unsupported attempts that waste paid tokens.

Error Messages/Logs

The only error surfaced was a single line, with no detail:

◇ ultraplan
Starting Claude Code on the web…
ultraplan: session creation failed — Repo is too large to teleport

No further logs were shown to the user. Notably absent (part of the problem):
- the repo's actual size,
- the teleport size limit,
- any docs link,
- any mention that a GitHub remote is also required.

A second, separate failure mode (the bare-keyword trigger) produced no error at all — /ultraplan simply began launching a cloud session from the word appearing in a prompt:

◇ ultraplan
Starting Claude Code on the web…

Steps to Reproduce

ultraplan

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.183 Claude code

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_