Conciseness

The body is dense and assumes competence — 'The constructors are discoverable (dir(rrb), help(rrb.Spatial3DView)); this skill covers what you can't guess' — with no padding or explanation of concepts Claude already knows. Nearly every line carries non-obvious, project-specific detail (mode semantics, reference-integrity rules, store-identity binding), matching the 'every token earns its place' anchor.

Actionability

Mostly executable: complete, runnable examples for blueprint construction, the ViewerClient screenshot loop, dataset schema enumeration, and concrete MCP call shapes ('rerun_set_blueprint with {"table": {}, "mode": "partial"}'). It falls short of 5 because the TableBlueprint snippet references undefined variables ('blueprint.save("my_app", path)', 'dataset.register_blueprint(path.as_uri(), ...)') without showing how path or dataset are obtained — a minor but real copy-paste gap.

Workflow Clarity

A clear numbered sequence (1. Read the data → 5. Iterate from screenshots) with explicit validation checkpoints — 'verify state before assessing screenshots', 'After each write, check the views and their reports in rerun_get_state, plus logs' — and an error-recovery feedback loop ('If the structure differs, inspect rerun_get_blueprint before retrying') plus a final screenshot check before reporting completion. This matches the top anchor including the retry loop.

Progressive Disclosure

Good structure with clearly signaled one-level external references (MCP docs for component overrides and table blueprints, the table-layout howto, and a 'See also' section for sibling skills), and no bundle files exist so nothing is misfiled. It is not 5 because a single ~240-line file inlines fairly deep reference-grade material — table column path-escaping rules and the named-coordinate-frame troubleshooting — that would fit a reference file under a leaner overview.

Specificity

The description lists multiple concrete actions — 'Change how the Rerun viewer looks by writing a blueprint', 'laying out a recording or dataset', 'hiding columns, changing column order or filtering rows' — and enumerates topic coverage (archetype-to-view mapping, rrb construction API, contents grammar, table blueprints, screenshot loop). This matches the anchor for comprehensive coverage of specific concrete actions; it is not 4 because both recording-layout and table-column use cases are covered with actionable verbs.

Completeness

It explicitly answers both questions: what ('Change how the Rerun viewer looks by writing a blueprint, for a recording or for a table') and when ('Read this whenever asked to change the look of the viewer, such as...') with concrete trigger phrases. This is a direct match for the top anchor; the 4 anchor's 'when could be more explicit' does not apply given the enumerated examples.

Trigger Term Quality

Natural phrases users would say are well covered: 'change the look of the viewer', 'laying out a recording or dataset', 'hiding columns', 'changing column order', 'filtering rows'. It falls short of the 5 anchor because a few natural asks are missing as triggers — e.g. changing colors/colormaps (a whole body section) and 'arrange the panels/views' — but coverage is clearly above the midpoint.

Distinctiveness Conflict Risk

The 'Rerun viewer' blueprint niche is clear and the triggers are scoped to visual appearance and table layout, keeping it distinct from sibling skills like rerun-data-model or rerun-catalog-queries. Minimal conflict risk matches the 5 anchor; it is not 4 because the scope boundary is explicit rather than merely implied.