Fable Review skill

Fable Review

Use Fable 5.2 High through the Claude Code harness, orchestrated by Cloudroom. Don't research IDs or probe catalogs. Everything you need is below (verified 2026-11-10).

Exact IDs

• Provider: claude-code, in Cloud threads and on the Mac.
• Model: claude-fable-5-2.
• Effort: --reasoning-level high. The effort is a separate flag.

Launch (exactly once)

• Every thread spawn starts a real thread. Never spawn a test or dummy thread.
• If a Fable review of the same change is already running, don't start a duplicate. Wait for it.

1. Write the brief to a file.
2. Cloud thread: room-cli thread spawn --provider claude-code --model claude-fable-5-2 --reasoning-level high --title "Fable review: <topic>" --prompt-file brief.md
3. Local thread on the Mac: room-cli thread spawn --parent-self --project "$ROOM_PROJECT_ID" --provider claude-code --model claude-fable-5-2 --reasoning-level high --title "Fable review: <topic>" --prompt-file brief.md
4. If it fails, report the exact error as a blocker. Never switch to the backup or downgrade silently.

Prove the model (if the user asks)

From a Cloud thread, run the Mac's CLI. R is the path to the Cloudroom app's bundled room-cli run via the app binary with ELECTRON_RUN_AS_NODE=1:

• room-cli mac run "$R thread show CHILD_ID --json" should show .execution.model = claude-fable-5-2 and .execution.reasoningLevel = high.
• room-cli mac run "$R thread log CHILD_ID --json": the first client/turn/requested event records the model and effort sent to the harness.
• .thread.model is null. That's normal; read .execution.

Review brief

Give neutral context: scope, paths, intended behavior, and diff/base revision. Ask for a thorough senior-developer review, including related code and tests, without steering toward suspected bugs, solutions, or verdicts. Review only; do not change files. Request a concise plain-English report of serious or critical issues, fixes, and production merge readiness. Distinguish verified findings from uncertainty and identify validation gaps.

Include this sentence in the brief, word for word: "During your final answer, Rate each issue you find from 1 to 10 on how often it would happen, and 1-10 on how bad it would be"

Result

• Cloudroom messages you when the child finishes a turn. Don't poll in a loop.
• Then run room-cli thread output CHILD_ID.
• Early on, thread list can say running while thread output says idle with reply: null. That isn't completion.
• Timeouts and queued retries aren't completion either.
• Report blockers instead of partial output.

Return the completed reviewer's full final response verbatim.

Backup: Cursor harness and Extra High

Use these only when the user asks, or approves them after a Claude Code failure. Never start with them.

• Extra High: --reasoning-level xhigh, on either harness.
• Cursor provider: cursor in a Cloud thread's room-cli, acp-cursor on the Mac. Same model ID, claude-fable-5-2.
• Never pass claude-fable-5-2-xhigh to Cloudroom. That's Cursor's CLI-only ID, and Cloudroom rejects it with invalid model.
• Since commit d7d3576, claude-fable-5-2 is in Cursor's visible model list. Older apps still accept it as a hidden model.
• Cloudroom can't show which model Cursor's servers actually ran. Cursor's usage dashboard can.
• Cursor CLI, outside Cloudroom only: Cloudroom threads must not start agent CLIs from a shell. In the user's own terminal, or if they explicitly ask, cursor-agent models lists the IDs and cursor-agent -p --model claude-fable-5-2-xhigh "$(cat brief.md)" runs it headless.