Skip to content
Build with Frank

The artifacts behind every build in the lab.

AGENTS.md. Repo conventions. Prompt packs. PR review checklists. Eval harness. Content flywheel. Forkable, documented, voice-spec clean. One year of updates. Commercial license.

Request early accessSee it in use

Early-access pricing while v1 ships. List price €149 at general availability.

Inside the pack

Five core artifacts. Each one tested in production.

These are the exact files Frank uses on every build in the public lab. They have shipped real product surfaces — not slideware.

AGENTS.md

The first file every agent reads when it joins a repo. Defines roles, scope, escalation paths, banned operations, review gates. Drop-in for Claude Code, Codex, Antigravity, Cursor.

Repo conventions

Folder structure, branch strategy, commit message patterns, and the PR review checklist used on every build in the lab.

Prompt packs

Tool-specific reusable prompts: planning, refactor, test repair, content extraction, build-log entry. Markdown, forkable, voice-spec clean.

PR review checklist

A reviewable checklist for agent-authored PRs: scope, tests, security, voice, banned operations. Plus a Claude Code skill to run it automatically.

Evaluation harness

A minimal evaluation scaffold to verify agent output against acceptance criteria before merge. Pluggable across LangGraph, Claude Agent SDK, OpenAI Agents.

What you actually get

A real repo, not a PDF.

Clone, fork, ship. Every artifact is plain Markdown or TypeScript. Zero proprietary format, zero lock-in.

Commercial license · One year of updates · 14-day refund

  • AGENTS.md template (Markdown, MIT-style commercial license)
  • Repo conventions doc and folder scaffold
  • 12 reusable prompts across planning, refactor, content
  • PR review checklist + Claude Code skill to run it
  • Minimal eval harness (TypeScript + Python flavors)
  • Build-log MDX template (the one used on /agentic-builder-lab)
  • LinkedIn post template (9 hook × close combinations)
  • One year of updates as new builds add new artifacts

Preview

A peek at AGENTS.md.

The file every agent reads first. Below is an abbreviated excerpt — the full version ships in the pack.

# AGENTS.md

## Roles
- Architect (human): owns the brief, the acceptance criteria, the merge
- Builder agent: writes the code, runs the tests, opens the PR
- Reviewer agent: runs the PR checklist, posts findings as comments
- Verifier (human): final review before merge

## Scope
- Builder may touch: app/, components/, content/, lib/
- Builder may NOT touch: .env, secrets/, infra/, .github/workflows/

## Banned operations
- No force push, no destructive git operations
- No package additions without a one-line rationale in the PR body
- No deletion of files outside the current task scope

## Escalation
If the builder cannot complete the task in 3 turns, open a draft PR
with current state and a question for the human.

## Review gates
1. All tests pass locally
2. PR checklist comment is green
3. No banned operations in the diff
4. Human verifier approves

FAQ

What buyers ask before clicking.

What format is the pack delivered in?

A GitHub repo you can clone, plus a downloadable .zip. Every artifact is plain Markdown or TypeScript — no proprietary format, no lock-in.

License and support?

Commercial license for use across your projects (including client work). One year of updates included — every new artifact added to the pack lands in your version. Email support for setup questions.

What tools does it support?

Claude Code, Codex, Antigravity 2.0, Cursor, Windsurf. AGENTS.md and the prompt packs are tool-agnostic; the Claude Code skill is opt-in.

How is it different from a free GitHub gist?

The pack is the result of shipping 8 production builds with these exact files — every artifact has been used, broken, and refined in real work. The eval harness alone saves a week of scaffolding.

Refund policy?

14-day no-questions refund. Email hello@frankx.ai.

Stop scaffolding. Start shipping.

The pack saves the week you would spend writing AGENTS.md, building an eval harness, and refining prompts. Skip that week.

Request early accessWorkshop or sprint instead

Or book a 20-min intro and we figure out whether the pack alone is enough.