
prod-check
Claude Code skill: audit your app before you take money from customers. 180 checks, 3 tiers, evidence-backed.
Install with your AI
Paste into Claude Code, Cursor, or any agent — it reads the repo and wires the tool into your project.
Install and set up prod-check (claude-skill project) into my current project. Found on https://claudeers.com/prod-check Repo: https://github.com/lisbeth718/prod-check Homepage/docs: — Detected install method: claude-skill → # copy this skill into .claude/skills/prod-check/ Category: integrations. Platforms: api, web. Read the repo's README for exact setup and env vars, then install it and wire it into my project. Claudeers Health Verdict: active; community-verified: false. Confirm the source before running anything.
# copy the skill dir into your project: # .claude/skills/prod-check/ (or ~/.claude/skills/prod-check/ for all projects)
git clone https://github.com/lisbeth718/prod-check
// compatibility
| Platforms | api, web |
|---|---|
| Operating systems | — |
| AI compatibility | claude |
| License | MIT |
| Pricing | open-source |
| Language | — |
prod-check
Find the bugs that end startups before your first customer finds them: leaked keys, other users' data, fake payments, no restorable backup.
Run /prod-check in your repo. In about 30 minutes your agent checks up to 68 high-risk items with evidence (file:line, command output); anything it cannot reach is reported UNKNOWN. It then hands you the top 5 fixes.
Read-only in production; replays write only to a test environment you provide; test alerts and messages are sent only with your OK. Built for agent-written apps: catches pasted keys, packages that don't exist, IDOR in server actions, Supabase tables open to the browser, and webhooks nobody verifies.
Going enterprise later? full covers all checks up to SSO, SOC 2 and DPAs. You don't need those yet.
Based on a checklist by Mari Luukkainen.
Install
Claude Code:
git clone https://github.com/lisbeth718/prod-check ~/.claude/skills/prod-check
Then run /prod-check in your app's repo.
Other agents: give the agent the path to SKILL.md and tell it to follow it. It reads the files under references/ as needed.
Humans: work through CHECKLIST.md. Each line has an ID, a tier and a gate. references/checks/ says how to verify each one.
Usage
/prod-check quick audit (default), 68 checks
/prod-check full every check in scope, 187 checks
/prod-check quick auth billing only those sections
/prod-check full 05-database-data-model
Quick mode runs and reports only a short list of the highest-risk checks. See references/quick-profile.md.
Fixes are proposed after the report and applied only when you say go.
How it works
Tiers and gates
A weekend landing page and a multi-tenant billing app do not need the same checks. Each check has a tier, and some sections are gated on a feature.
- T0 Public site: no accounts, no payments, no customer data.
- T1 Paying: any of: accounts, customer data, payments.
- T2 Enterprise/regulated: SSO, questionnaires, contract terms, or regulated data (health, financial account, legal, member, union or political data). Ordinary PII such as CVs and emails stays T1.
- A (advisory): reported as notes, never a FAIL.
The auditor checks everything at or below your tier. Anything above is N/A, not FAIL.
Gates are features you may not have: accounts, multi-tenant, payments, usage billing, uploads, imports, transactional email,
background jobs, API or webhooks, AI, URL fetching, impersonation, subscriptions, multiple locales, health data, card data, database, self-hosted servers. A gated-off section is N/A.
multi-tenant means an org, team or workspace entity where several users share records; per-user ownership alone is not multi-tenant.
Some things only a person can answer (on-call, postmortems, pen tests, who owns the domain). The auditor asks you once. An artifact it can verify is PASS, your word alone is ATTESTED, and an admitted gap is FAIL.
Stages
T0 and T1 rows carry a stage; T2 and advisory rows use -. Stages: launch (before the first real user outside the team or the first invoice, whichever comes first), 10 (before 10 customers) or 100 (before 100 customers). All T0 rows are launch.
Quick mode is the high-risk T0 rows already in the quick list plus every T1 launch row; stages tell you what to fix first, not what is in scope.
Missing IDs were retired and are not reused.
Review
For T1 and T2 apps a second agent that did not run the audit re-checks every FAIL, every PASS marked "(code only)", plus every PASS in AUTH-, AUTHZ-, WEB-01 to WEB-03, BILL-01 and BILL-02. If no reviewer can be spawned, the report says "Reviewer: none" and lists those rows.
Access
The repo, a shell on the box, read-only database access, and the cloud consoles (hosting, DNS/CDN, email provider, billing). Anything the auditor cannot reach is UNKNOWN, stated up front in the report header as "Access gaps".
The agent will see secrets
An agent with this access will read env files, configs and logs. Run it with evidence hygiene on: env vars listed by name only,
no env/printenv/docker inspect dumps, no secret values pasted into the report, redact on sight. Use read-only credentials and a test environment for replays.
The report
One table: section, ID, check, tier, stage, status (PASS, FAIL, ATTESTED, UNKNOWN, N/A), evidence, fix, effort.
FAILs are ordered by data exposure, data loss, money, availability, then the rest. The top five fixes come last.
See examples/report-t1-saas.md (a fictional app) and references/report-template.md.
Add your stack
The checks are vendor-neutral. Commands for a concrete stack live in references/stacks/.
Copy references/stacks/_template.md, fill it in, and open a PR.
example-nextjs-postgres-hetzner-cloudflare-ses.md is a filled-in example.
Contributing
- One PR per check (or per stack file).
- Every check needs an ID, a tier, a concrete way to verify it, and the evidence to record.
- Keep checks vendor-neutral; put vendor commands in an "e.g." clause or a stack file.
- Process-only checks are marked
attestand need a question inreferences/attestation.md. - Keep
CHECKLIST.mdin sync with the check files. No em dashes.
License
MIT. See LICENSE.
// faq
What is prod-check?
Claude Code skill: audit your app before you take money from customers. 180 checks, 3 tiers, evidence-backed.. It is open-source on GitHub.
Is prod-check free to use?
prod-check is open-source under the MIT license, so it is free to use.
What category does prod-check belong to?
prod-check is listed under integrations in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/prod-check)
// retro hit counter
[](https://claudeers.com/prod-check)
// reviews
// guestbook
// related in Integrations & Connectors
Use claude code and codex for free in the terminal, VSCode extension, and discord like OpenClaw (voice supported)
Bridge local AI coding agents (Claude Code, Cursor, Gemini CLI, Codex) to messaging platforms (Feishu/Lark, DingTalk, Slack, Telegram, Discord, LINE, WeChat…
All parts of Claude Code's system prompt, 27 builtin tool descriptions, sub agent prompts (Plan/Explore/Task), utility prompts (CLAUDE.md, compact, statusli…
Claude Code skill/plugin: immersive scroll-scrubbed 'fly through the world' landing pages generated with Higgsfield (Emons-style diorama flights, seamless co…