rule

running tests

How to run tests, check for failures, and rerun specific failing tests — use the :agent variants of slow commands

prisma/prisma-next/blob/HEAD/.agents/rules/running-tests.mdcofficiallistedunreviewednot running here yet

Set up

Skills are installed by placing their folder where your agent loads skills.

Part of the collection prisma/prisma-next.

Copy the text of running-tests.mdc into the instructions file your agent reads (CLAUDE.md, AGENTS.md, .cursor/rules, …). Written for AI coding agents in general (from where the project keeps it; we have not run it).

README

synced Oct 14, 2026

How to run tests, check for failures, and get specific failures

1. Run a test suite

Each slow command has an :agent variant. The first line of its output prints the exact log path:

$ pnpm test:packages:agent
log: wip/test-packages.20260615-190441.32317.log

The full output goes to that file. The exit code is written to a sibling .exit file (same path with .log → .exit). Use the path printed by the command — don't guess one. Every run gets its own timestamped path, so history is preserved automatically. Contents of wip/ are gitignored.

Each :agent run has a hard timeout (default 5 minutes, override with AGENT_CMD_TIMEOUT_SECONDS — bump it for a legitimately long suite). If the run hangs — e.g. vitest freezes after printing its summary — the child process group is killed and the .exit file records 124 rather than blocking forever.

Command	Purpose
`pnpm test:packages:agent`	Unit tests for every package
`pnpm test:integration:agent`	Integration suite (PGlite + mongodb-memory-server)
`pnpm test:e2e:agent`	End-to-end suite
`pnpm build:agent`	Build every package (turbo)
`pnpm typecheck:agent`	Typecheck every package (turbo, cached)
`pnpm lint:agent`	Biome lint
`pnpm lint:deps:agent`	Dependency-layering check
`pnpm fixtures:check:agent`	Regenerate fixtures and diff them

2. Run one package's tests

pnpm --filter @prisma-next/cli test

Runs only that package's own test script — seconds, not minutes. Output goes to your terminal; no capture needed.

3. Rerun a specific failing test

By file:

pnpm --filter @prisma-next/cli test test/commands/migration-list-json-golden.test.ts

By test name within a file (vitest's -t filter):

pnpm --filter @prisma-next/cli test test/commands/migration-list-json-golden.test.ts -t "pins head ref decoration"

4. Find specific failures in the captured output

Use the log path the :agent command printed (the shell exit status tells you pass/fail; the .exit sibling file records the same number). If non-zero, list failing files and assertions:

grep -nE " FAIL |Test Files" wip/test-packages.20260615-190441.32317.log
grep -B2 -A20 "FAIL "         wip/test-packages.20260615-190441.32317.log

If exit is non-zero but the log has no FAIL lines, the suite crashed before any test ran — read the tail:

tail -50 wip/test-packages.20260615-190441.32317.log

Don't

Don't run the full suite when you already know which file is failing. Use the targeted commands in §3 — they run in seconds.
Don't run the bare canonical commands (pnpm test:packages, etc.) for a full-suite run. They dump the entire output into your context window. Use the :agent variant.
Don't pipe a :agent command to tail/grep/head. The point is that the output is already on disk — read the log path the :agent command printed.
Don't re-run a suite to refine a grep. Grep the file you already have.
Don't pass --force to turbo. The cache is correct by default. Only force when this same run edited tsconfig.json or a lockfile.