Aller au contenu principal

Skill: dev-prisma

Inline

Pointer to Prisma's own agent skills (prisma-cli, prisma-client-api...) and how to install them, plus the schema-change workflow. Use for Prisma ORM work (schema, migrations, queries, schema.prisma) only when no prisma-* skill is installed; when one is, use it instead.

Configuration​

PropertyValue
Contextinline
Pre-approved toolsNone declared: the project's permission settings apply
Triggerkeywords: dev, prisma, prisma —

Detailed description​

Prisma ORM (pointer + workflow)

Prisma's own skills (prisma-cli, prisma-client-api, …) track each Prisma release: install them. This pointer adds the workflow below.

Prisma publishes them at prisma/skills (9 skills: CLI, Client API, database setup, Prisma Postgres, the v6 → v7 upgrade, driver adapters, MongoDB, Compute). The foundation's former 418-line skill drifted on every Prisma release.

git clone https://github.com/prisma/skills ~/dev/vendor-skills/prisma
git -C ~/dev/vendor-skills/prisma checkout 1123817e60d15ca0f3af91878923241dee7e3b09   # the registry pin
ln -s ~/dev/vendor-skills/prisma/prisma-cli ./.claude/skills/prisma-cli
ln -s ~/dev/vendor-skills/prisma/prisma-client-api ./.claude/skills/prisma-client-api

Recipe entry: docs/recipes/recommended-vendor-skills.md §"Prisma — prisma/skills".

Schema change workflow​

  1. Edit schema.prisma: the relation on both sides; indexes per .claude/rules/prisma.md.
  2. npx prisma migrate dev --create-only --name <change> — writes the migration without applying it.
  3. Read the generated SQL: a rename shows up as drop + add (data loss) — split it as .claude/rules/prisma.md says before anything runs.
  4. npx prisma migrate dev to apply it, then npx prisma generate — since Prisma 7, migrate dev no longer regenerates the client (v6 did). Commit the migration folder with the schema; one migration per change, never a second init.
  5. Production: apply the committed migrations from CI or the deploy step — the exact commands, and what never runs against production, are in .claude/rules/prisma.md.

Discipline that holds whichever skill fires​

The migration, secret and query rules live in .claude/rules/prisma.md, scoped to the Prisma files, so they load even when a vendor skill fires instead of this one.

  • Security: never fetch passwordHash or any sensitive column without explicit need; select over include — .claude/rules/prisma.md and .claude/rules/security.md.
  • TDD with a real DB: integration tests hit a real test database (Docker Compose pattern), never a Prisma mock — the dev-tdd skill.
  • Postgres interop: with Supabase, Prisma runs against the same Postgres and RLS still applies — the dev-supabase skill.

Why this pointer steps aside when Prisma's skills are installed, even though a session then often loads no skill: measured 2026-10-04, the work came out right either way (schema and migration correct 3/3 with and without this skill firing), and .claude/rules/prisma.md loads from the files in both cases — eval/skill-triggering/FINDINGS.md.

Automatic triggering​

This skill is automatically activated when:

  • The matching keywords are detected in the conversation
  • The task context matches the skill's domain

Triggering examples​

  • "I want to dev..."
  • "I want to prisma..."
  • "I want to prisma — ..."

Context inline​

Inline means the skill runs in the conversation (the default, no context key):

  • Sees the full history
  • Its work is visible as it happens
  • Right for almost every skill

See also​