Conciseness

Most of the ~880-line body is genuinely non-obvious qsv-specific knowledge (flag semantics, refusal modes, precedence rules) rather than concepts Claude already knows, so it is not anchor-2 padding — but several passages are padded or repeated: the geocode-enabled-build vs --geocode-flag distinction is explained three times (Stage 3a bullets, the probe script's printed output, Guardrails), the x-qsv.tour "version must stay the integer 1" rule appears in both Stage 2.5's table and Guardrails, and the denominator derivation story is told twice. This lands on anchor 3: mostly efficient but with unnecessary explanation that could be tightened — notably by deduplicating the geo-routing lore and trimming issue-number asides (#4401, #4528, #4547).

Actionability

The guidance is overwhelmingly executable — exact qsv invocations with flags, parameter expansions like ${GEOJSON:+--geojson "$GEOJSON"}, complete copy-paste python heredocs for the dictionary coverage check and GeoJSON key ranking, and specified AskUserQuestion prompts with defaults. It falls short of 5 on real gaps: Stage 2.5's centerpiece `edit_dictionary.py` is described as living "beside this SKILL.md" but no such file exists in the bundle, and the Stage 2 endpoint resolution block leaves $MODEL unresolved until an interactive prompt Claude must fabricate structure for.

Workflow Clarity

Stages 0–6 are explicitly ordered ("The order is load-bearing. Clean first, then describe, then draw.") with rationale for the ordering, and validation is woven throughout: Stage 0 preconditions, Stage 1's report-before-apply with a rows_affected-vs-nullcount sanity check, the Stage 2 dictionary coverage check gated before Stage 4 spend, Stage 5's verify-then-report with an explicit "Never claim success without checking", and Stage 6's capped two-loop refine→re-render→re-verify cycle. This matches the anchor-5 example's validate→fix→retry structure with feedback loops at every risky step.

Progressive Disclosure

The stage-based headers make the document navigable, but it is a single ~880-line monolith with no reference files at all: the seven-row x-qsv hints table, the ~90-line GeoJSON feature-id ranking script, and the Stage 6 tour-refinement protocol are exactly the material the rubric expects split into one-level-deep referenced files. Worse, the one bundle file the body does reference — edit_dictionary.py "beside this SKILL.md" — is absent, so the reference is dangling in the bundle as shipped. Anchor 3 fits: real structure, but content that should be separate is inline and the sole script reference is not actually present.

Specificity

The description lists several concrete actions — "Build a Data Schematic", "driven by an LLM-inferred JSON Schema data dictionary", "browsable in-page next to the charts", "Optionally bins rows into GeoJSON regions" — which sits between anchor 4 (several specific actions, minor gaps) and anchor 5. It falls short of 5 because the actions are phrased through qsv-internal coinage ("Data Schematic", "qsv viz smart") rather than plainly named operations, and coverage of the pipeline (clean → infer → render) is implicit rather than listed.

Completeness

Both halves are explicit: what — "Build a Data Schematic with its Data Dictionary beside it — an interactive qsv viz smart dashboard driven by an LLM-inferred JSON Schema data dictionary" — and when — a full "Use when the user asks for..." clause with four concrete trigger phrasings plus the combined-goal case. This matches the anchor-5 good example structure exactly; nothing below 5 fits since neither half is missing or vague.

Trigger Term Quality

"Use when the user asks for a Data Schematic, a visual data dictionary, a documented dashboard or a dictionary-driven dashboard, or wants to explore and document a CSV" gives good keyword coverage including the core noun phrases plus "CSV" and "GeoJSON". It is not 5 because common natural variations are missing — e.g. ".csv" extension forms, "data dictionary", "document a dataset", "data profiling" — and most triggers are the skill's own coined terms rather than phrases an uninitiated user would spontaneously say.

Distinctiveness Conflict Risk

The niche is distinct — a dictionary-driven qsv dashboard with GeoJSON binning is unlikely to be the intended skill for generic charting or data-cleaning requests, and the triggers are mostly coined, unambiguous terms. It is not 5 because "documented dashboard" and "explore and document a CSV" are broad enough to overlap with generic data-documentation or dashboarding skills.