Ship-Safe · the pre-deploy gate for AI-built code

Open source · MIT

The pre-deploy gate for AI-built code.

AI writes code at machine speed and leaves quiet risks: a leaked key, a dirty working tree, a build that never compiled. Run one command in your repo and get a plain-language GO or NO-GO before you deploy. It reads and reports. Nothing leaves your machine unless you opt in to reporting.

Run it in your repo, no install. the six commands that matter:

your-repo · real commands, copy and run

npx getadvantage

you geta read-only GO or NO-GO for your repo before you deploy: dirty-tree guard, secret scan, tracked .env check, typecheck, plus three maps of what your app exposes. Add --build for the full production build.

npx getadvantage brief

you getPROJECT-BRIEF.md written from the real repository, so every model and every session reads the same ground truth first.

npx getadvantage handoff

you getHANDOFF.md with what changed since the last save point, plus the start prompt for your next session.

npx getadvantage fan-out 3

you getthree parallel worktree lanes on fresh branches, each wired to read the project brief before it works.

npx getadvantage fan-in

you geta collision map and a merge dry run. With --apply, a lane lands only while the combined tree stays green; a lane that breaks it is rolled back, never shipped.

npx getadvantage mcp

you geta dependency-free MCP server over stdio: six tools so your agent can read the brief, save a handoff and run the gate itself. No keys, no network.

Wire it into Claude Code:claude mcp add getadvantage -- npx getadvantage mcp

Or add it to your project:

npm install --save-dev getadvantage

View on GitHub →The package on npm

Node 18+. Dependency-free (Node built-ins only). Read-only, except the files you ask it to write.

What this page gives you

One command in your repo, a plain GO or NO-GO before you deploy, and the seven checks behind it.

One run

What a run looks like. Plainly.

Every check reports pass, warn or fail, then one overall verdict. A single blocking fail turns the whole run into a NO-GO. Here a broken build stops the ship, and the missing project brief is the one warning left to eyeball.

$ npx getadvantage --build

┌──────────────────────────────────────────┐
│  Ship-Safe — is this safe to ship?        │
└──────────────────────────────────────────┘

Checks
  ✓ Dirty-tree guard — Working tree is clean — nothing unintended would ship.
  ✓ Secret scan — Scanned 11 tracked/staged file(s) — no leaked-secret patterns matched.
  ✓ Tracked .env file — No .env files tracked by git (gitignored local .env files are fine and are never read).
  ✓ Typecheck (tsc --noEmit) — TypeScript compiled with no type errors.
  ✗ Production build (npm run build) — The production build failed — fix it before shipping.
      > acme-storefront@0.1.0 build
      > node scripts/build.mjs
      Bundling acme-storefront for production…
      Build failed: entry module app/main.ts not found (referenced from build config).
  ✓ Schema-bump check — Skipped — no app/lib/server/db.ts with a SCHEMA_VERSION sentinel in this project.

Overview — what your app has (read-only)
  ✓ API surface map — 3 route(s) · 2 look gated (session or cron secret) · 1 mutate (write) · 0 mutate without any obvious gate.
      /api/contact  [POST]  auth-gated
      /api/cron/digest  [GET]  secret-gated (cron/shared)
      /api/health  [GET]  public (read-only)
  ✓ Agents & integrations map — 1 integration(s) detected.
      OpenAI — key: OPENAI_API_KEY  (1 file)
  ✓ Schedules & jobs map — 1 job(s) · 1 scheduled in vercel.json · 0 ungated · 0 cron route(s) not wired to a schedule.
      /api/cron/digest  [0 8 * * *]  gated (CRON_SECRET)
  ⚠ Project brief — No project brief at PROJECT-BRIEF.md.
      project brief is stale, run `ship-safe brief` to refresh.

Verdict
  ✓ 8   ⚠ 1   ✗ 1

  NO-GO — 1 blocking issue(s). Do not ship until these are clear.

A clean run means these specific checks found nothing at that moment, not that the app is secure. The gate reports what it checked. It never calls your app secure.

Two ways from here: run the gate yourself, free and open source. Or we run the whole delivery line for you, gated like this on every change.

Run it yourself Book a call →

What it catches

Seven checks, each for a specific failure AI-built code ships often.

The safety checks own the NO-GO. The maps warn, never block.

Dirty-tree guardBLOCKSexplainer →A production deploy ships the working tree, not a tidy commit. This blocks the moment a tracked file is modified or staged, so no unrelated or half-finished work rides along by accident. The single biggest foot-gun in AI-built repos.
Secret scanBLOCKSexplainer →Reads every committed and staged file against eighteen leaked-secret patterns: OpenAI, Anthropic, Stripe live and restricted keys, AWS access keys, GitHub tokens, Google OAuth, Slack, SendGrid, npm access tokens, bare JWTs, database URLs with embedded passwords, private-key blocks, webhook and Vercel tokens. A match prints a masked fingerprint, never the secret, and blocks the ship.
Tracked .env fileBLOCKSexplainer →A committed .env is a leak by itself, whatever it contains. Git history keeps every value it ever held. Blocks on any tracked .env file; a local .env that is not gitignored warns, because it is one git add away.
TypecheckBLOCKSexplainer →Runs tsc --noEmit. Type errors block. A change that does not compile does not ship.
Production buildBLOCKSexplainer →With --build, runs the full npm run build. The build must complete for the change to pass. This is the check that has to read Compiled successfully.
Schema-bump checkWARNSexplainer →Additive database changes are skipped on a live database whose version already matches, so a schema change that forgets to bump the version passes on a fresh database and silently no-ops in production. This warns loudly when that happens.
Envelope mapsWARNSexplainer →Three read-only maps of what the app actually exposes: every route and whether it authenticates its caller, every external and LLM integration and the key behind it, every scheduled job and whether it is gated. A write route with no gate, a secret reachable from the browser bundle, or a publicly triggerable cron surfaces here. Informational, never a block on its own.

The same gate, in every engagement

This is the exact gate every change clears before we ship it for you. When we build and ship your AI, the same checks run on every change, and each piece of work ends on a plain go or no-go, in writing. The gate is open so you can read exactly what we run. Our full method, from the controlled zone to the proof record, is on the methodology page.

Evaluating this for a larger organization? The enterprise picture, including what we are not, is on the enterprise page.

Want AI built and shipped, gated like this? Start with a call.

Thirty minutes over video, directly with the founders. No obligation.

Book a call →How we deliver