[zantvoort]

Summary

The skills under website/static/skills/ are instructions a coding LLM loads before writing Storm code, and the changelog is read by people scanning for what changed. Both now read for that audience.

• No database names. No skill names a database product or a vendor feature by its vendor. A rule says what the database does: a default collation that ignores case, a list bound as one array parameter (id = ANY(?)), the in-memory database @StormTest provides by default, the database's accent-removing function wrapped in a function its index accepts. Per-database tables for JSON types, JSON aggregates and key generation are single generic statements. @StormTest(database = ...) is shown with the constant for the project's database. Dependency coordinates such as st.orm:storm-postgresql and com.h2database:h2 stay, since they are how a dependency is declared.
• Concise, rule first. Each rule opens with one bold imperative sentence, followed by its reason and a minimal example, with lists in place of paragraphs. Facts stated several times in a file are stated once. No paragraph exceeds 600 characters, down from about seventy that did. Every rule, API name, example and cross-reference is kept; the Java and Kotlin twins state the same rules in the same order.
• Changelog. The unreleased entries are one or two sentences each, with the identifiers a user searches for and the issue number. The longest is 272 characters.
• Corrections found on the way. The migration skill told agents never to add a generated folded column, which contradicted Query columns: metamodel fields for columns a query names but entities do not carry #743 and Query column: name the function that computed it, so a compared value is folded by the database in the same statement #745; it now describes that as the conditional step, with the query column naming its function, and points to the search documentation for the statement per database. A wrong operator example and a template placeholder typo in the Kotlin query skill, two stale comments, a duplicated item number in the entity skills and a vendor date function in a code-first example are corrected.

The branch is stacked on #746, which touches the same skill lines; the base moves to main when that merges.

Test plan

• Case-insensitive scan of every skill for database product names: hits only inside dependency coordinates and one container image name
• No em dashes, no paragraph over 600 characters, none of the banned words, in any skill or the changelog section
• Every skill keeps its headings and item numbering; the Java and Kotlin twins were diffed section by section

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!