Prompt Forge

One brief → production-grade prompts in RISEN, CO-STAR, RTF, CRISPE, or BAB. The model analyses the brief, asks targeted gap questions, generates against a hard section contract, validates the result, and lets you fine-tune with plain-English feedback.

Live: https://risen-forge.vercel.app

Features

• Personalised onboarding. First visit asks role (11 personas), goals, which AI tool you use and experience level. Profile is stored on the account (or in the browser for guests) and re-openable from the header.

• Simple / Advanced modes. Simple shows just the essentials in plain language; Advanced exposes target, framework, autonomy, chain, few-shot and extensions. Defaults from the profile.

• Role template library. 60 ready-made briefs (marketing campaigns and ads, EA emails and minutes, study plans, investor updates, sales sequences, lesson plans, job ads, research synthesis, engineering tasks…), your goals first, every role browsable.

• Framework registry with auto-pick. Inferred from task type and target (Writing → CO-STAR, Strategy → CRISPE, chat refactor → BAB, chat research/data → RTF, else RISEN). Can be overridden.

• Clarify step. The model scores brief completeness and asks up to 4 high-leverage questions.

• Rules with reasons. Each constraint carries a "because…", which models follow far more reliably than bare rules.

• Few-shot slots. Up to 3 input→output pairs, rendered verbatim as <example> blocks.

• Chain mode. Produces 2–4 sequenced prompts with explicit hand-off contracts instead of one mega-prompt.

• API target extras. Assistant prefill and a JSON output schema.

• Agentic extensions. Acceptance criteria, evidence gates, authority boundaries, Definition of Done, and self-critique.

• Contract validation. Required headings are checked after generation, with one automatic repair pass if any are missing.

• Offline template mode. Deterministic output for every framework with no API call.

• Light/dark/system themes. Accessible custom dropdown and version history (v1, v2…), with .md export.

Accounts, plans & data

• Auth: Auth.js v5 (Google / GitHub) with the Prisma adapter; providers turn on when their env vars exist.
• Database: Prisma 7 + Postgres (Vercel Prisma Postgres). Migrations run automatically at build when DATABASE_URL is set; without it the app runs in guest-only mode.
• Plans (lib/plans.ts): Guest 3 AI generations/day · Free 30/month · Pro 1,000/month. Autofill and analysis are free (burst-limited). ADMIN_EMAILS get Pro.
• History: last 5 versions per prompt plus up to 10 pinned; auto-fixes replace the attempt they fix. Synced for signed-in users, in the browser for guests.

Security model

The server is the security boundary; client-side checks are UX only.

Layer	Control
Request	Zod schema on every body, unknown keys dropped, per-field length caps, 120 KB body cap, same-origin check
Text hygiene	NFKC normalisation; zero-width, bidi-override, Unicode tag and control characters stripped (client + server)
Prompt injection	Security preamble with highest priority; all user content wrapped in <untrusted_*> tags; spoofed delimiter/system tags defanged; suspicious phrasing logged
Output	Per-request canary detects system-prompt leaks (output discarded); off-task output (no framework sections) discarded; JSON outputs re-sanitised
Abuse / cost	DB-backed burst limits and plan quotas per user or hashed IP; free auto-repairs capped relative to paid calls; provider details never exposed
Data	Ownership checked on every project/version route; raw IPs never stored
Headers	CSP, X-Frame-Options: DENY, nosniff, strict referrer, permissions policy

Stack

Next.js 16 (App Router) · React 19 · TypeScript · Prisma 7 · Auth.js v5 · Zod · any OpenAI-compatible LLM API · Vercel

app/
  page.tsx                     # brief → clarify → forge/refine
  Onboarding.tsx, useProfile.ts # role/goals/tool/level setup
  Account.tsx, useHistory.ts   # account bar, library, versions
  api/generate/route.ts        # analyze | generate | refine | suggest (guarded, metered, streamed)
  api/me, api/profile, api/projects/**  # account, profile, saved prompts/versions
  pricing, signin, terms, privacy
lib/
  forge.ts                     # frameworks, contracts, validator, offline compiler
  personas.ts, profile.ts      # personas, templates, profile schema
  server/{db,account,guard,metering,projects}.ts
prisma/schema.prisma           # User/Account/Session, Project, Version, UsageEvent

Run locally

npm install
cp .env.example .env.local     # AI provider, DATABASE_URL, AUTH_* vars
npx prisma migrate deploy
npm run dev

Environment

See .env.example for every variable with notes.

License

MIT