Conciseness

The body is dense and gotcha-focused, assuming Claude's competence ('Tiling/compression alone does not establish COG layout', 'matching array shape alone does not prove alignment', 'Cloud filters at scene level do not replace per-pixel masks') with no re-explanation of known concepts. It is 4 rather than 5 because of minor trimmable material: the verbatim citation block, three separate uv pip install lines, and pinned version strings ('GeoPandas 1.2.0, Rasterio 1.5.2...') that add time-sensitive detail, though these are at least quarantined in a dedicated 'Tested scope and installation' section.

Actionability

Every section ships complete, executable code: the raster recipes call bundled functions whose signatures (write_ndvi, terrain_metrics, classify_imagery) match scripts/raster_workflows.py exactly, and the vector, STAC, OSMnx, and COG snippets are copy-paste ready with real arguments and inline error handling ('if not items: raise ValueError', 'if not valid: raise ValueError(errors)'). The body explicitly flags file-dependent examples as templates, covering the common cases honestly.

Workflow Clarity

The numbered 6-step workflow has explicit verification checkpoints ('Verify CRS from authoritative metadata. ... Never infer an unknown CRS from plausible bounds', 'Check numeric expectations on a small known fixture before scaling up') and the code raises on failure (missing CRS, no scenes, invalid COG). It falls short of 5 because there is no fix-and-retry feedback loop: failures are raised but no recovery guidance follows, leaving the loop implicit.

Progressive Disclosure

The body is an overview with quick-start recipes inline and all deeper material split across 15 real, one-level-deep reference files, each listed with a one-line descriptor in the References section and contextually linked where relevant (machine-learning.md in the classification section, data-sources.md and remote-sensing.md in the catalogs section). The bundled script's import requirements are documented and all referenced paths verified to exist.

Specificity

The description enumerates several concrete capability areas and actions ('processing satellite imagery, aligning coordinate systems and raster grids, accessing STAC catalogs, analyzing geospatial time series') alongside named domains (remote sensing, vector and raster GIS, spatial statistics, terrain and network analysis). It sits at 4 rather than 5 because the coverage is domain enumeration with some action phrasing rather than a fully comprehensive list of concrete actions, and terms like mapping or visualization of spatial data are absent.

Completeness

Both questions are explicitly answered: the 'what' is stated concretely ('Supports geospatial research workflows for remote sensing, vector and raster GIS, spatial statistics, terrain and network analysis, and machine learning for Earth observation') and the 'when' is an explicit clause with five concrete trigger scenarios ('Use when processing satellite imagery, aligning coordinate systems and raster grids, accessing STAC catalogs, analyzing geospatial time series, or implementing scientific GIS workflows in...'). Voice is third person, so no penalty applies.

Trigger Term Quality

Natural user phrases are well covered: 'satellite imagery', 'coordinate systems', 'raster', 'GIS', 'STAC catalogs', 'geospatial time series', 'remote sensing', 'terrain', plus eight programming-language names. It falls short of 5 because common synonyms and file extensions users would naturally say (e.g. 'map', 'shapefile', '.shp', 'GeoTIFF', 'DEM') are missing, though coverage clearly exceeds the 'some relevant keywords' level of 3.

Distinctiveness Conflict Risk

The skill occupies a clear geospatial/Earth-observation niche with distinct triggers (STAC catalogs, raster grids, satellite imagery, terrain analysis) that are unlikely to fire for unrelated skills. Overlap risk with generic data-analysis skills is minimal, matching the clear-niche anchor.