docs(skills): improve accuracy, structure, and consistency across CLI skills

[eddietejeda]

Summary

A quality pass over all four agent skills (hotdata, hotdata-search, hotdata-analytics, hotdata-geospatial), driven by a per-skill audit and verified against the built CLI. Also folds in the recent-change doc updates: #170 (unscoped indexes list now includes managed databases) and #167 (KEDA cold-start hint).

Net ~184 fewer lines with higher accuracy and signal-to-noise.

Accuracy fixes (verified against --help)

• Removed the nonexistent query status --output flag (core + analytics).
• Added update to the core top-level subcommand list.
• Documented results list --limit (default 100, max 1000) vs queries list (20) — agents were likely under-fetching.
• Split indexes list (filters optional) from indexes delete (requires --connection-id + --schema + --table + --name).
• Standardized --catalog as a catalog alias; fixed Chain naming (uses --table/--catalog, there is no chain-table --name); added the missing embedding-providers create --output flag.

Structure / progressive disclosure

• hotdata-geospatial (285 → 166 lines): added the missing hotdata query execution layer (the skill never showed how to run its SQL) + a table-qualification/quoting note, sharpened the area-conversion caveat (cos(latitude) shrink), and moved the function catalog + unit tables to a new references/functions.md.
• core hotdata (373 → 299 lines): collapsed the ~5×-repeated context/DATAMODEL material into one section, de-bolded prose, and dropped trailing "Workflow:" blocks already covered by references/WORKFLOWS.md.

Consistency

• Unified context scoping language as database-scoped (was "workspace") across the core reference files.

#170 / #167 doc updates (carried in this branch)

• indexes list (no flags) now spans regular connections and managed databases (shown under the internal __db_<id>.<schema>.<table> label); documented in search SKILL, INDEXES.md, MODEL_BUILD.md, and the data-model template.
• Note that search --table on a managed DB needs the SQL catalog prefix (default.public.<table>), not the __db_… label or a connection id.
• Added a "Cold starts (worker wake-up)" note so agents don't treat the ~10–20s first-call pause as a hang.

Verification

• Every hotdata command referenced across the skills maps to a real subcommand.
• Spot-checked changed flags against the built CLI (query, query status, databases load, indexes delete, embedding-providers create, results list).
• All references/… links resolve to existing files.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[claude]

Reviewed against the CLI source in src/command.rs. All accuracy claims check out: query status has no --output flag, update is a real top-level subcommand, results list defaults to 100/max 1000 vs queries list 20, embedding-providers create has --output, indexes list filters are optional, and the #database-context-named-markdown anchor resolves. Clean docs quality pass.