Conciseness

Efficient and platform-specific throughout — exact path rules ('/Users/<user>/.cursor/projects/<workspace>/canvases/<name>.canvas.tsx'), import restrictions ('Import only from cursor/canvas'), and labeling specs — with only minor trims available ('A canvas exists to show real content', 'Not everything deserves equal treatment', the squint-test aside). Not 5 because a few sentences restate design fundamentals Claude already knows; not 3 because there is no padded section.

Actionability

Highly concrete guidance: exact directory and filename rules, 'Default-export the top-level component', 'No fetch(), no network calls', label examples ('API error rate by service', 'Latency (ms)'), a copy-paste markdown link format, and a pointer to the SDK .d.ts declarations for exact exports. Not 5 because there is no minimal .canvas.tsx component skeleton to start from; not 3 because every step is executable as written.

Workflow Clarity

Explicitly ordered ('Follow the workflow below in order') from the decide-then-write sequence through the Pre-delivery self-check, with real feedback loops: the 'Canvas TypeScript check' line is designated the authoritative diagnostics signal and Troubleshooting gives recovery steps for a blank canvas. Not 4 because both the validation checklist and the error-recovery loop are explicit rather than implicit.

Progressive Disclosure

Well-organized single-file structure with clear section headers and an on-demand, one-level-deep external reference ('~/.cursor/skills-cursor/canvas/sdk/index.d.ts — read them when you need exact exports'); no bundle files exist to inline or bury. Not 5 because the ~90-line body keeps all content inline with no split of design/troubleshooting guidance into separate files and the under-50-lines exception does not apply; not 3 because nothing that belongs in a separate file is buried.

Specificity

Names the domain ('A Cursor Canvas is a live React app that the user can open beside the chat') and concrete actions like 'render it in a rich canvas rather than dumping it into a markdown table' and 'read this skill whenever you create, edit, or debug any .canvas.tsx file', but the long enumeration ('quantitative analyses, billing investigations, security audits...') lists trigger scenarios rather than capabilities, and the pervasive second-person voice ('You MUST use a canvas', 'If you catch yourself') costs a point per the rubric's voice rule — base 4, penalized to 3. Not 2 because the domain and several concrete behaviors are explicitly named.

Completeness

Explicitly answers both: what ('A Cursor Canvas is a live React app that the user can open beside the chat') and when ('You MUST use a canvas when the agent produces a standalone analytical artifact', 'whenever you create, edit, or debug any .canvas.tsx file') with concrete trigger phrases. Not 4 because the 'when' clause is as explicit and enumerated as the anchor-5 example.

Trigger Term Quality

Covers natural user phrasing broadly — 'quantitative analyses', 'billing investigations', 'security audits', 'timelines', 'charts', 'tables', 'MCP tools', 'markdown table' — and includes the file-extension trigger '.canvas.tsx' plus the synonym 'canvas'. Not 4 because coverage including synonyms and the file extension is comprehensive rather than merely good.

Distinctiveness Conflict Risk

Mostly distinct — the '.canvas.tsx' file type and Cursor-specific locations give it a clear niche — but 'any response that benefits from visual layout' and 'charts, tables' create minor overlap risk with general visualization or formatting skills. Not 5 because of that breadth; not 3 because the core trigger is a proprietary artifact few other skills would claim.