.github

Imported from qabdlab/.github (AGENTS.md). Install upstream with npx skills add qabdlab/.github. Copyright stays with the author.

Working approach

The agent's role in this project is advisory only. The maintainer writes the code and configs himself; the agent's tasks are limited to:

• Hints, advice, and explanations of how and why things are done
• Reminders about missed steps, inconsistencies, and potential problems
• Reviewing/re-checking the project state when asked
• Writing commit message texts

Everything the agent produces is reviewed by the maintainer before being applied. Do not edit files, commit, or push unless explicitly asked to.

Commits

Follow Conventional Commits:

• Format: type(scope): summary — imperative mood, lowercase, no trailing period, ≤72 chars
• Types: feat, fix, refactor, perf, test, docs, build, ci, chore
• Scope is optional; use the adapter/module name when it clarifies (feat(sqlite): ...)
• Breaking changes: ! after the type/scope and a BREAKING CHANGE: footer

The body is required for anything non-trivial and must explain what was done and why:

• The motivation/problem the change solves, not a restatement of the diff
• Key decisions made and alternatives that were rejected, if any
• Side effects, behavior changes, and anything reviewers/future readers need to know
• Wrap body lines at ~72 chars; write in English

One logical change per commit — don't mix refactoring with features or formatting with logic.

Tests

• Use bun test; test files live in tests/ (integration) or co-located as src/**/*.test.ts (unit)
• Name tests after observable behavior, not implementation details (returns null for SQL NULL, not test #3)
• Structure: arrange → act → assert; one behavior per test
• Test the public adapter API (what Prisma calls), not private helpers
• Integration tests run against real databases: in-memory bun:sqlite for SQLite; Postgres/MySQL via connection URLs from .env.test (never hardcoded)
• Tests must be independent and order-agnostic: no shared mutable state, clean up created data/tables (beforeEach/afterEach)
• Cover edge cases of type conversion (NULL, BigInt, dates, bytes, JSON) — that's where adapter bugs live
• Coverage is collected on every run; don't chase 100%, do cover every branch of value mapping

Code

• Formatting and linting are biome's job (bun run check:fix) — don't hand-format
• TypeScript strict mode; avoid any (biome warns), prefer precise types from @prisma/driver-adapter-utils
• Public API entry points get JSDoc; internals stay comment-light — comment only non-obvious constraints

Bun

Default to using Bun instead of Node.js:

• bun <file> instead of node/ts-node, bun test, bun install, bunx <pkg>
• Bun automatically loads .env — don't use dotenv
• bun:sqlite for SQLite, Bun.sql for Postgres/MySQL — never better-sqlite3, pg, postgres.js, or mysql2
• Bun.file over node:fs readFile/writeFile; Bun.$ for shell commands
• Full API docs: node_modules/bun-types/docs/**.mdx