Prisma Skills Pack FAQ: Nine Agent Skills, One ORM
Frequently Asked Questions
What is the Prisma skills pack and what ships in it?
Why do all nine skills show nearly identical install counts?
Is prisma-cli the ORM skill or the Platform CLI skill?
prisma (this skill, version 7.9.1). The public-beta Platform package is @prisma/cli with binary prisma-cli, routed to the prisma-compute skill. The file spells it out: use prisma-compute for Compute apps and workspace auth, prisma-postgres for Platform projects and databases, and this skill for ORM commands like init, generate, migrate, db push/pull, dev, debug, and studio.What actually breaks when upgrading Prisma to v7?
prisma-client with required output paths and generated entrypoints (client, browser, models, enums), driver adapters become mandatory for SQL providers, config moves to prisma.config.ts with explicit environment loading, and the module system goes ESM-first with a CommonJS fallback via moduleFormat = ‘cjs’. It also lists removed features — middleware, metrics, and legacy CLI behavior — plus a special section for Accelerate users. The skill’s crash case is import errors right after upgrade, which trace back to the new generated entrypoints.Does the v7 upgrade guide apply to MongoDB projects?
What runtime prerequisites do the skills assume?
bunx --bun prisma so Prisma uses the Bun runtime instead of silently falling back to Node. The client-api skill (v7.9.1) organizes its reference by priority — model queries, transactions, and raw SQL are CRITICAL; client construction, query shape, filters, and relations are HIGH.Which skill should I install first?
What does the postgres platform skill cover that database-setup doesn't?
prisma postgres link, managing resources through the Management API or its typed SDK, and connection choices between direct TCP and serverless drivers. prisma-database-setup stays in ORM-land — provider strings, env-var configuration, and client instantiation. The boundary is the same flavour as cli-vs-compute: platform vs ORM, and the pack keeps the two deliberately separate.Why a Prisma Skills Pack FAQ Exists#
Prisma shipped a full agent-skills pack without much fanfare, and most walkthroughs still treat it as one skill. It’s nine. The pack covers the ORM CLI, the client API, per-database setup, the v6→v7 migration, MongoDB’s unique dead-end, two hosted-postgres skills, app deployment, and custom driver adapters — with some genuinely sharp boundaries drawn between them. I wrote these answers after pulling the SKILL.md files out of prisma/skills and reading them line by line, alongside the 2026-09-01 skills.sh install snapshot. Every claim below traces to a file in that repo, not to a README summary.
The Short Answers#
- One pack, nine skills. Install them together; they share the same version train and the same workflow-first philosophy.
- The numbers are a pack counter. 235K–253K with lockstep weekly waves means “the pack got installed”, not “client-api is more popular than cli”.
- The cli skill is the ORM. The Platform’s
@prisma/clibelongs to prisma-compute — the boundary warning in the file is there because people keep confusing the two. - v7’s breaking changes are named. Driver adapters required, prisma-client generator, prisma.config.ts, ESM-first with a CommonJS escape hatch.
- MongoDB has no v7 path. v6 is terminal, Prisma Next is the successor; an agent that says “upgrade MongoDB to v7” is hallucinating.
What the Leaderboard Numbers Actually Tell You#
The 2026-09-01 scrape put the pack’s nine skills between 235,757 and 253,040 installs — a tighter cluster than any other family on the board, with weekly curves moving in lockstep. Step back and the signature is unmistakable: this is the same pack-counter pattern we documented for azure-skills in the azure-messaging intro and azure-diagnostics intro — install counts measure bundle adoption, not individual skill popularity. The prisma pack’s curve shape differs from azure’s in one way worth noting: azure rides a long plateau, while prisma’s weekly wave drops from a ~43K peak to ~20K over four weeks, which fits a newer pack burning off initial installs. Trust the files for per-skill signal, the totals for pack-level reality.
Common Mistakes to Avoid#
Mistake #1: Treating prisma-cli as the platform CLI. The SKILL.md boundary section exists because this one bites constantly. ORM commands live in prisma-cli; @prisma/cli app deployment lives in prisma-compute; hosted-database work lives in prisma-postgres.
Mistake #2: Running v7 upgrades on MongoDB projects. The upgrade skill explicitly disclaims itself, and v7 has no connector. Route MongoDB through prisma-mongodb-upgrade and its stay-or-migrate decision table instead.
Mistake #3: Assuming install totals rank individual skills. A pack that moves as one unit on the leaderboard rewards you with no per-skill heat map. Judge prisma skills by their file contents, versions, and your actual workflow.
Mistake #4: Forgetting the Bun flag. On Bun runtimes, bunx prisma without --bun silently uses the Node runtime. The skill’s bunx --bun prisma convention is a one-word fix for a whole class of subtle runtime mismatch bugs.
Start With the CLI Pair#
Grab prisma-cli and prisma-database-setup first — that pair covers six of your ten daily ORM interactions. npx skills add https://github.com/prisma/skills pulls the pack, then let your agent read the SKILL.md files before touching a schema. If you’re sitting on a v6 codebase, add prisma-upgrade-v7 to the install and read its removed-features list before you run a single migration command — it names exactly which flags and middleware will vanish. And when a teammate asks whether “skills” are just prompts, point them at the how to create your own AI agent skills guide so they can see how much structure a real skill pack carries.
Related Articles
I Ran npx create-db and Got a Live Postgres in Seconds
Narrative of a real create-db run: a throwaway Prisma Postgres appeared in seconds with no signup, honored a 30-minute TTL, and answered SQL over TLS.
My Spec Session With to-Spec: No Interview, Just Synthesis
I ran Matt Pocock's to-spec skill on a half-baked feature and it refused to interview me: pure synthesis, one test seam pinned, then a ready-for-agent spec.
Resolve Git Merge Conflicts With AI: The 5-Step Workflow
A step-by-step guide to Matt Pocock's resolving-merge-conflicts skill, the five-step workflow that bans --abort, keeps both intents, and finishes every rebase.
Domain Modeling With AI: Build a CONTEXT.md Glossary
A step-by-step domain modeling walkthrough with Matt Pocock's skill — run the five moves, write a CONTEXT.md glossary, offer ADRs when all three tests pass.