Content
65%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.
A strong, information-dense PostgreSQL design reference with genuinely actionable DDL and well-chosen gotchas, held back by its shape rather than its substance: it is a flat topic-by-topic reference with no sequenced design workflow or validation checkpoints, and no progressive disclosure into reference files despite covering ~15 distinct topic areas in one file. Tightening the longest bullets and splitting extensions/JSONB detail into references would lift the two weakest dimensions.
Suggestions
Add a short sequenced design workflow (e.g., define access patterns → choose types → apply Core Rules → add indexes/constraints → test DDL in a transaction and roll back) with an explicit validation checkpoint, which would also satisfy the database-operations feedback-loop requirement.
Move the Extensions list and the JSONB guidance (opclass trade-offs, generated-column extraction) into a references/ file and link them one level deep from SKILL.md, keeping the core rules and examples inline.
Split the overloaded Strings bullet (TOAST strategies, collations, CITEXT) into separate short entries or a reference, and complete fragmentary snippets like the RLS policy example.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rule-oriented ('Heap storage: no clustered PK by default... CLUSTER is one-off reorganization'), assuming Claude already knows PostgreSQL basics rather than explaining them. A few entries run long (the Strings bullet packs TOAST strategies, collations, and CITEXT into one paragraph) and could be trimmed, keeping it below the 'lean, every token earns its place' anchor. | 4 / 5 |
Actionability | Most guidance is executable — concrete DDL like 'CREATE INDEX ON tbl (user_id) WHERE status = 'active'' and complete copy-paste example tables (users, orders, profiles). Some entries remain fragmentary ('FOR VALUES IN ('us-east', 'us-west')', 'CREATE POLICY user_access ON orders ... USING (user_id = current_user_id())' referencing an undefined function), so it stops short of fully copy-paste-ready coverage. | 4 / 5 |
Workflow Clarity | This is reference material organized by topic, not a sequenced design workflow — there is no 'analyze access patterns → define types → add constraints → index → validate' order. The rubric's scoring note caps database-operation skills without explicit validate/fix/retry feedback loops at 3; 'Safe Schema Evolution' offers transactional-DDL testing ('BEGIN; ALTER TABLE...; ROLLBACK;') but no explicit validation checkpoints, so 3 is the ceiling and also the best fit. | 3 / 5 |
Progressive Disclosure | Sections are clearly headed (Core Rules, Data Types, Indexing, Partitioning, Examples), but the entire ~200-line reference lives in SKILL.md with no bundle files and no offloading — the Extensions list, JSONB deep-dive, and TOAST/collation details are exactly the material the anchor-3 example describes as inline content that could live in separate reference files. | 3 / 5 |
Total | 14 / 20 Passed |