Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is exceptionally actionable and safely structured for destructive database operations, with concrete commands, ready SQL examples, and meaningful validation checkpoints. Its main weakness is the inlined ~230-line table catalog that inflates token cost and would be better split into a reference file, especially since the live `schema` output is already declared authoritative.
Suggestions
Move the Core Tables catalog to a references/ file (e.g., references/tables.md) and keep a brief purpose/write-boundary summary in SKILL.md — the live `schema` result is already authoritative, so full column listings need not occupy the main context window.
Add a general post-write verification step to the Workflow (e.g., re-query the affected rows after `write` to confirm the intended change), extending the repair-case verify pattern to all writes.
Trim per-table entries to purpose and write boundary, dropping full column listings for non-critical tables and letting `schema <table>` supply columns on demand.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The operational sections (Commands, Workflow, Safety Rules, Common Queries) are lean and efficient, but the ~230-line Core Tables catalog inlines full column lists for every table even though the file itself says "Always treat the live `schema <table>` result as authoritative", so the bulk of the body duplicates retrievable runtime output. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready guidance throughout: concrete subcommands (`python scripts/mp-db.py tables`, `schema downloadhistory`, `write "UPDATE subscribe SET state = 'S' WHERE id = 123"`), eight ready-to-run SQL examples, a dialect-comparison table, and troubleshooting steps. | 5 / 5 |
Workflow Clarity | A clear five-step workflow with inspect-before-write sequencing ("run `tables` first, then `schema <table>`") and real checkpoints (confirm destructive writes, suggest backups, enforce `WHERE`, report affected row count, "verify the managed API can read the repaired value" for repairs); the minor gap is that post-write re-verification is only specified for the systemconfig repair case, not the general write flow. | 4 / 5 |
Progressive Disclosure | The bundle is a single correctly-referenced script (`scripts/mp-db.py`) with no references/ directory, and sections are clearly headed, but the huge inlined table catalog matches the anchor-3 pattern of "200 lines of API reference that could be in a separate file" — it belongs in a one-level-deep reference file rather than the main SKILL.md. | 3 / 5 |
Total | 15 / 20 Passed |