Ultrareview v2.1.96: docs describe /tasks command + claude ultrareview --json subcommand that don't exist; findings hard to read after completion

Summary

The Ultrareview (/code-review ultra / /ultrareview) docs at https://code.claude.com/docs/en/ultrareview describe two output/management surfaces that don't exist in shipped v2.1.96:

1. /tasks slash command — docs say "Use /tasks to see running and completed reviews, open the detail view for a review, or stop a review that is in progress." In v2.1.96 there is no such slash command; /tasks is not in /help's output.
2. claude ultrareview standalone subcommand with --json flag — docs say "Use the claude ultrareview subcommand to start an ultrareview from CI or a script…" with a --json flag for raw bugs.json. In v2.1.96 there is no ultrareview subcommand; running claude ultrareview 707 parses ultrareview as the prompt and --json errors with error: unknown option '--json'.

Combined, these two gaps make it hard to read a finished Ultrareview's findings.

Reproduction

Tested on Claude Code 2.1.96 (output of claude --version).

$ claude --version
2.1.96 (Claude Code)

$ claude ultrareview 707 --json
error: unknown option '--json'

$ # Inside Claude Code interactive session:
/tasks
# Slash command not found.

The slash command /code-review ultra <PR> and its alias /ultrareview <PR> do work — they launch a cloud sandbox session, display a panel with Setup → Find → Verify → Dedupe phases, and end with "Review complete — N issues found" and a list of red-text finding previews. The previews are truncated and not directly readable (one-line ellipsis e.g. "Bucket arrays from $alloc free-list reuse contai…").

What "Review complete" surfaces (or doesn't)

After "Review complete — 3 issues found" the docs say "the verified findings appear as a notification in your session. Each finding includes the file location and an explanation of the issue so you can ask Claude to fix it directly."

Observed v2.1.96 behaviour: no notification arrived in the local Claude Code session. The full text of the findings was only readable by opening the cloud session URL (https://claude.ai/code/session_<id>?from=cli) in a browser — which the launch banner does print, but it's framed as the "Track" link for live progress, not the canonical "see your findings" surface. The browser page itself has the file:line locations and finding titles visible directly (no truncation), but no input field for further interaction — the session is read-only after completion.

A spurious task-notification event also landed mid-run with status: failed, summary: remote session exceeded 30 minutes — but the panel continued past that point and completed normally. The session ID in that event matched the still-running review. May or may not be related; flagging in case it's useful.

Suggested fixes (in priority order)

1. When the cloud review completes, surface the full finding text as a session message — the docs' described "notification" channel. Whatever broke that delivery in v2.1.96 vs the docs' described version.
2. Make the red-text finding previews in the panel clickable to expand into the full text. The previews truncate at the panel width with no visible affordance for "click to see more."
3. Either ship the claude ultrareview subcommand the docs describe, or update the docs to remove it. Same for /tasks.
4. Treat the browser cloud-session page as a documented output surface, not just a "Track" link. The launch banner could say e.g. "View findings: <url>" in addition to the "Track" framing.

Environment

• Claude Code: 2.1.96
• macOS: 15.7.3 (Apple Silicon)
• Login: Max plan, Claude.ai auth (not API key)
• Triggered via /ultrareview 707 (the alias form)
• Cloud session: completed successfully, 3 issues found+verified+deduped

The substantive findings themselves were excellent (correctly identified two real GC bugs not caught by the parallel CodeRabbit review — verified and fixed). The feedback here is purely about the find-text-after-completion UX in v2.1.96, not about review quality.