Conciseness

The body is dense and assumes competence — no basics of SQL, DataFrames, or kernels are explained — with every section carrying operational content. It sits at 4 rather than 5 because a few rationale passages could be trimmed, e.g. 'Reason: each one-shot process pays roughly a second of spawn-plus-import overhead...' and 'One scan up front converts every later query from a file re-scan into milliseconds', which justify rather than instruct. It is well above the 'mostly efficient but some unnecessary explanation' middle anchor because the padding is minor and interleaved with actionable rules.

Actionability

Concrete, runnable commands appear throughout: 'run scripts/ensure-js-deps.sh once', 'uv run --with ...', 'CREATE TABLE t AS SELECT ...', "FROM 'data.csv'", 'uv run scripts/quick-query.py <file> [SQL] (--filter, --describe)', and '.pl() via Arrow'. It stops short of 5 because some guidance is directional rather than copy-paste ready — 'Dynamic-import it, connect once, query across cells' and 'sys.path.insert the printed directory and import' omit the actual code, deferring it to references.

Workflow Clarity

A clear decision sequence is present — choose a surface, choose an engine, 'Probe before you compute — one cell: file size, free RAM, and (when unclear) a row count', place the work, then the output contract ('report row counts and timing... then stop'). Checkpoints exist (the sizing probe, timing/row-count reporting, the mandated visual check in the visualization reference), matching 'clear sequence with most checkpoints present'. It is not 5 because error-recovery feedback loops are not explicit in the body, and not 3 since the sequence and checkpoints are concrete and well-ordered rather than implicit.

Progressive Disclosure

SKILL.md is a genuine overview with all detail pushed to five real reference files (all present in references/) and five scripts (all present in scripts/), each inline pointer paired with a 'Read X before Y' condition and consolidated in a References table with a 'When' column. References are one level deep — reference files point only to scripts and one sibling reference, never to further nested documents — matching the top anchor for clear, well-signaled navigation.

Specificity

The description enumerates multiple concrete operations — 'group-by/join/aggregation, time series, distributions, cleaning, or plotting a dataset' — alongside the engines used ('DuckDB, Polars'), giving comprehensive coverage of the data-analysis domain rather than generic language. It is not score 4 because no meaningful capability in its stated scope is left uncovered; it is not below because every verb names a real, distinguishable action.

Completeness

Both halves are explicit: the what ('Processes and analyzes data with resident-kernel engines (DuckDB, Polars) and one-shot tools') and an explicit when clause with concrete triggers ('Use for CSV/parquet/JSON analysis, group-by/join/aggregation, time series, distributions, cleaning, or plotting a dataset'). This mirrors the score-5 example structure exactly; the 'when' is neither missing nor merely implied, so 3 and 4 anchors do not fit.

Trigger Term Quality

It uses natural user phrasing including file formats/extensions ('CSV/parquet/JSON') and the words users actually say: 'group-by', 'join', 'aggregation', 'time series', 'distributions', 'cleaning', 'plotting a dataset'. Coverage including synonyms and extensions matches the top anchor; at most a variant like 'statistics' is absent, which does not drop it to the 'a few natural terms missing' level.

Distinctiveness Conflict Risk

The DuckDB/Polars tabular-analysis niche is clear with distinct triggers, but 'plotting a dataset' creates minor overlap with dedicated visualization/charting skills, matching 'mostly distinct; minor overlap risk with closely related skills'. It is not 5 because of that charting overlap, and not 3 since the engine-specific, file-format-driven triggers are far more specific than 'works with document files'.