Conciseness

The ~36-line body contains zero concept explanations or padding — every line is a command, a path, or an operational rule ("Reload after state-heavy changes", "Never type into an existing session"). It assumes Claude's competence and every token earns its place, matching the 'lean and efficient' anchor exactly.

Actionability

The Commands block is fully executable, copy-paste-ready bash (`node $S new-session`, `node $S wait-idle 180`, `node $S shot /tmp/after.png`, `node $S eval '...'`) verified against the real scripts/cdp.mjs subcommands, and the sqlite3 fallback includes a complete query string. It covers the common cases for this skill end-to-end.

Workflow Clarity

The sequence (find window → export CDP_URL → new-session → send → wait-idle → shot/eval) is clear with inline ordering comments ("always start a fresh session first", "until the Run task button is back") and wait-idle acts as a synchronization checkpoint before screenshotting. It falls short of a 5 because there is no error-recovery guidance — nothing on what to do if wait-idle times out, no OpenWork page is found, or the send fails.

Progressive Disclosure

The skill is under 50 lines with no need for external reference files, and its sections (Find the window / Commands / Rules) are well-organized and easy to navigate. The single bundle file, scripts/cdp.mjs, is referenced by its exact working path, matching the actual bundle structure.

Specificity

The description lists multiple specific concrete actions — "Evaluate JS, take screenshots, open a new session, send a prompt with a chosen model, wait for the run to finish" — comprehensively covering the script's actual subcommands, in third-person voice. It matches the top anchor and is not merely 'several actions with minor gaps'.

Completeness

It explicitly answers both what ("Drive a running OpenWork desktop window over CDP from the shell" plus the five concrete actions) and when ("Use when checking a UI change by hand in a world or pnpm dev, or when reproducing a chat state") with concrete trigger phrases, exactly matching the top anchor.

Trigger Term Quality

Natural trigger phrases are present ("checking a UI change by hand", "reproducing a chat state", "screenshots", "send a prompt"), giving good keyword coverage. It falls below the comprehensive anchor because "CDP" is unexpanded jargon a user might not say, and common phrasings like "screenshot the desktop app" or "debug the UI" variants are missing.

Distinctiveness Conflict Risk

The niche is a specific app's desktop window driven over CDP — "OpenWork desktop window over CDP" — with triggers (manual UI check, chat-state reproduction) distinct from automated eval/spec testing referenced in the body. Minimal conflict risk with other skills.