Prototype to production

Hardens a working prototype with a prioritized list of fixes instead of a rewrite - error handling, input validation, env config, logging, core-flow tests, auth checks, pinned dependencies and deploy notes. Use when a demo or vibe-coded app needs to go live.

Vibe codingPack: Vibe Coding69 linesprototype-to-production

Install

Run one command — it saves SKILL.md into your agent's skills folder. Restart the agent and the skill is picked up automatically.

Claude Code — macOS, Linux
mkdir -p ~/.claude/skills/prototype-to-production && curl -fsSL https://anymodel.org/en/skills/vibe-coding/prototype-to-production/SKILL.md -o ~/.claude/skills/prototype-to-production/SKILL.md
Codex, Cursor — macOS, Linux
mkdir -p ~/.agents/skills/prototype-to-production && curl -fsSL https://anymodel.org/en/skills/vibe-coding/prototype-to-production/SKILL.md -o ~/.agents/skills/prototype-to-production/SKILL.md
Claude Code — Windows PowerShell
New-Item -ItemType Directory -Force "$HOME\.claude\skills\prototype-to-production" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/vibe-coding/prototype-to-production/SKILL.md -UseBasicParsing -OutFile "$HOME\.claude\skills\prototype-to-production\SKILL.md"
Codex, Cursor — Windows PowerShell
New-Item -ItemType Directory -Force "$HOME\.agents\skills\prototype-to-production" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/vibe-coding/prototype-to-production/SKILL.md -UseBasicParsing -OutFile "$HOME\.agents\skills\prototype-to-production\SKILL.md"

Any other agent: save the file into its skills folder, or give it the link to the file: https://anymodel.org/en/skills/vibe-coding/prototype-to-production/SKILL.md

SKILL.md

Raw

Prototype to Production

A prototype that works is an asset. Do not rewrite it. Find the gaps that can hurt real users, rank them, and close them in small commits, riskiest first.

Workflow

  1. Run the app and its tests once to learn the baseline. Note what the app does and what data it holds. Do not change anything yet.
  2. Audit against the checklist below. For each item record: present, missing or partial, with a file reference.
  3. Produce a prioritized list (format below) and show it to the user. Get agreement on the scope before editing; the user may not want all of it.
  4. Fix items one at a time, each as its own small change with a check and a commit. Keep behavior the same unless the item is a bug.
  5. Finish with deploy notes and a list of what you deliberately left out.

Audit checklist

Security and data (highest priority)

  • Secrets: no keys, tokens or passwords in code or git history; all read from environment variables; .env is ignored by git; a .env.example lists names only. If a secret was committed, tell the user to rotate it; do not rewrite history yourself.
  • Auth: every non-public route checks identity on the server, not just in the UI. Check that users can only read and change their own records (try another user's id).
  • Input validation: all request bodies, query params, file uploads and webhooks are validated for type, size and range at the boundary. Database queries use parameters, never string concatenation.
  • Output: user-provided text is escaped when rendered; no stack traces or internal errors returned to clients.
  • CORS, cookies (HttpOnly, Secure, SameSite) and rate limits on login and expensive endpoints.

Reliability

  • Error handling: external calls (network, database, third-party APIs) have timeouts and handled failures. No empty catch blocks. User-facing errors are clear.
  • Unhandled exceptions are caught at the top level and logged without crashing the whole process silently.
  • Data: migrations exist instead of ad-hoc schema edits; a backup plan exists for the database; destructive operations ask for confirmation.
  • Idempotency for payment, email and webhook handlers.

Operability

  • Config: ports, URLs, feature flags come from environment with sane defaults; fails fast with a clear message if a required variable is missing.
  • Logging: structured or at least consistent lines with timestamps and request ids; no secrets or personal data in logs; levels are used.
  • A health endpoint and a documented start command.

Quality

  • Tests for the core flows (sign up, the main create/read action, payment if present), not for every function. They run with one command.
  • Dependencies: lockfile committed, versions pinned, npm audit or pip-audit style check run and findings triaged.
  • Lint or type check passes; dead code and debug leftovers (console logs, test routes, default admin password) removed.

Priority format

P0 (before any real user)
1. <issue>  where: file:line  fix: <one line>  effort: S/M/L
P1 (first week)
...
P2 (later)
...
Left out on purpose: ...

P0 means a user could lose data, see someone else's data, or an attacker could take control.

Deploy notes to write

Required environment variables (names, purpose, example format), build and start commands, the port, persistent storage paths, how to run migrations, how to roll back to the previous version, and what to watch after release.

Pitfalls

  • Rewriting in a "better" framework. Out of scope unless the user asks.
  • Adding layers (queues, caches, microservices) with no measured need.
  • Fixing everything at once in one huge commit.
  • Breaking behavior while adding validation: run the main flow by hand after each change.
  • Running audit commands against a production database or real third-party accounts. Use local or test data and ask first.

Source: AnyModel skills catalog · License: CC BY 4.0

Connect through AnyModel API

This skill runs on whatever model your agent uses. Point Claude Code, Codex or Cursor at AnyModel to run it on Claude, GPT, Gemini and more with one key — pay per token, no subscription.

More from the pack Vibe Coding