Conciseness

Nearly every line is environment-specific and non-inferable ('$fig.auto-flushes at script end', 'GROUP has no fills/strokes/cornerRadius', 'up to 5 .screenshot() calls per tool call'), with no filler explaining concepts Claude already knows — better than anchor 3. It is not 5 because of repeated content: the hex helper appears twice (pattern block and Critical Rule 6), Critical Rule 9 duplicates the 'Node property gotchas' children list, and the 'componentKey from search_design_system' import point is restated three times.

Actionability

Copy-paste-ready, complete code for every common operation — the autoLayout-with-children builder, $fig.query().set() updates, instance creation with props, gradient stops, the hex helper, page switching — plus the 'full surface' API listing. All cited reference files (fig-builder.md, critical-rules-deep.md, variable-patterns.md, plugin-api-standalone.d.ts, etc.) exist in the bundle, matching anchor 5's 'fully executable, covers the common cases'.

Workflow Clarity

Multi-step processes are explicitly sequenced with validation checkpoints and error-recovery feedback loops: incremental placeholder+replace building with .screenshot() verification per step, the two-call variant-discovery pattern (inspect componentPropertyDefinitions, then instantiate), rehydrating variable handles across calls, and the 3-call bulk-mutation budget ending in 'FINAL REPORT... STOP'. Rule 12/15 ('unexpected error is server-side — change approach'; '3 retries → switch approach') close the retry loop, so the batch-operation cap does not apply.

Progressive Disclosure

The body points to a real, verified 14-file reference bundle one level deep, each entry labeled with its content ('WRONG/RIGHT examples, full node-type pitfall list', 'grep, don't read whole'), and a dedicated References section — matching anchor 4's 'good structure, references mostly clear'. It is not 5 because the '$fig create + mutate API (full surface)' section and parts of 'Node property gotchas' inline API-detail content that fig-builder.md and plugin-api-standalone.index.md already exist to carry, making the main file longer than a clean overview.

Specificity

The e.g. list names multiple concrete actions — 'create/edit/delete nodes, set up variables or tokens, build components and variants, modify auto-layout or fills, bind variables to properties, or inspect file structure programmatically' — which is anchor-5 coverage, but the description is written in second person ('you MUST invoke this skill', 'NEVER call use_figma directly'), invoking the rubric's -1 specificity penalty. It is not a 3 because the action list is comprehensive, not merely 1-2 actions.

Completeness

The 'when' is explicit and specific ('Trigger whenever the user wants to perform a write action or a unique read action that requires JavaScript execution in the Figma file context'), so the anchor-3 cap for missing trigger guidance does not apply. The 'what' is only conveyed indirectly through the e.g. list and the phrase 'requires JavaScript execution in the Figma file context' — the description never states in its own words what the skill does (execute JavaScript in Figma files via the Plugin API), which keeps it at 4 rather than 5.

Trigger Term Quality

Strong natural terms tied to the domain — 'nodes', 'variables or tokens', 'components and variants', 'auto-layout', 'fills', 'bind variables', 'inspect file structure' — matching what users asking for Figma edits would say. It falls short of anchor 5 because common synonyms like 'design', 'colors', 'styles', 'text styles', or 'icons' are absent, leaving a few natural entry phrases uncovered.

Distinctiveness Conflict Risk

Triggers are tightly bound to the `use_figma` tool and Figma-specific vocabulary ('Figma file context', 'auto-layout', 'components and variants'), giving a clear niche well above anchor 3. It is not 5 because phrases like 'build components and variants' overlap with sibling skills this body references (figma-generate-library, figma-generate-design), so a natural-language request could route here when a companion skill is the better first load.