AI Native Founder Skill Pack

中文说明

A Codex-ready workflow pack for founders building AI-native startups.

This pack turns startup discussion into structured founder workflows: diagnose the current stage, identify the evidence gap, choose the right next action, and only then decide whether to validate, build, launch, automate, or scale.

What This Pack Is For

AI tools make it easier to ship software, but that also makes it easier to build the wrong thing faster. This pack helps a founder keep the work stage-gated:

• Idea: prove the problem is real, specific, frequent, and painful.
• MVP: test whether a focused solution creates repeat use, revenue, referral, or strong qualitative pull.
• Launch: harden PMF evidence, product operations, reliability, security, and growth loops.
• Scale: build enterprise readiness, delegation, GTM systems, data loops, workflow lock-in, and defensible moat.

The core rule is:

Do not build until evidence justifies it.
Do not scale until operations can survive it.
Do not automate judgment before the decision rule is explicit.

Who It Is For

This pack is useful for:

• Solo founders and small teams building AI-native products.
• Founders using Codex, Claude Code, or other agentic coding tools.
• Early operators who need repeatable workflows for validation, MVP planning, PMF review, launch readiness, and scale preparation.
• Technical founders who want a forcing function against scope creep and premature building.
• Non-technical founders who want AI agents to help structure the startup process without replacing founder judgment.

It is not meant to be a generic prompt collection, a fundraising guide, or a substitute for customer conversations.

Source And Inspiration

This repository is inspired by Anthropic / Claude's official article The founder's playbook: Building an AI-native startup, published on May 14, 2026.

That playbook reframes the startup lifecycle for AI-native companies across Idea, MVP, Launch, and Scale. This repository adapts that lifecycle into a Codex-ready runtime pack: project instructions, Skills, custom Agents, shared templates, deterministic scripts, and examples.

This is an independent workflow pack. It is not an official Anthropic product.

Current Repository Structure

This repository is packaged in Codex runtime form:

.
├── AGENTS.md              # project rules Codex reads for this pack
├── README.md
├── README_CN.md
├── MANIFEST.md
├── .agents/
│   ├── skills/            # Codex-visible runtime Skills
│   └── shared/            # shared references, templates, and scripts
├── .codex/
│   ├── config.toml
│   └── agents/            # project-scoped Codex custom agents
├── docs/
├── examples/
└── assets/

AGENTS.md is the durable project instruction file. It is not a custom agent. It tells Codex how this pack should operate: keep stage gates, separate evidence from assumptions, protect safety boundaries, and route work through the right founder workflow.

How To Use It

Open Codex from the repository root:

codex

Start with the coordinator:

Use founder_orchestrator to diagnose whether this startup is at Idea, MVP,
Launch, or Scale. Separate evidence, assumptions, risks, and the next smallest
action.

A good founder intake prompt includes:

Product idea:
Target user and buyer:
Current product state:
Evidence gathered:
Usage, retention, revenue, or referral data:
Current bottleneck:
Constraints:

The expected flow is:

Current thread
  -> founder_orchestrator for stage diagnosis and routing
  -> specialist agent or Skill for focused work
  -> reviewer for adversarial challenge when needed
  -> current thread integrates the final decision

Skills

Skill	Purpose
founder-stage-diagnosis	Diagnose the evidence-backed stage and gate readiness.
idea-validation	Convert an idea into testable hypotheses and disconfirming evidence.
customer-discovery	Design interview targets, questions, probes, and synthesis.
mvp-scope	Define MVP boundaries, non-goals, and feature gates.
ai-coding-context	Produce AI coding context, architecture notes, and session logs.
pmf-feedback-loop	Interpret activation, retention, revenue, referral, and PMF signals.
founder-bottleneck-audit	Identify work to stop, delegate, automate, or keep founder-led.
launch-readiness	Audit PMF evidence, production readiness, security, ops, and growth.
scale-moat-system	Assess enterprise readiness, GTM systems, data loops, lock-in, and moat.

Agents

Agent	Use when
founder_orchestrator	You need stage diagnosis, routing, or final synthesis.
idea_validator	You need problem validation, customer discovery, or competitor threats.
mvp_architect	You need MVP scope, measurement guardrails, or coding-session context.
launch_operator	You need PMF review, launch readiness, product ops, or growth channels.
scale_operator	You need enterprise readiness, GTM systems, delegation, or moat analysis.
reviewer	You need an adversarial check for weak evidence, false PMF, or stage mismatch.

Common Prompts

Idea validation:

Use idea_validator to turn this idea into testable hypotheses, identify the
strongest disconfirming evidence, and design a customer discovery script. Then
use reviewer to challenge the result.

MVP planning:

Use mvp_architect to define the MVP scope, explicit non-goals, acceptance
criteria, measurement plan, and first AI coding session context. Do not write
code yet.

PMF review:

Use pmf-feedback-loop to evaluate these activation, retention, revenue,
referral, and Sean Ellis signals. Call out false positives and missing evidence.

Launch readiness:

Use launch_operator to audit PMF evidence, production readiness, technical debt,
security/compliance, product operations, growth channels, and founder
bottlenecks. Return PASS, HOLD, or REVIEW.

Scale and moat:

Use scale_operator to audit enterprise readiness, GTM systems, delegation,
workflow lock-in, proprietary data loops, and defensible moat. Use reviewer to
challenge the moat narrative.

Install Into Another Project

ZIP install, recommended

The simplest installation path works the same on Windows and macOS:

1. Open the repository on GitHub.
2. Click Code -> Download ZIP.
3. Extract the ZIP file.
4. Copy the extracted pack contents into your target project root.

If you publish a release ZIP whose files are already at the archive root, users can extract it directly into the target project directory.

If using GitHub's automatic Download ZIP, GitHub usually wraps the files in a top-level folder such as ai-native-founder-skill-pack-main/. Open that folder and copy its contents into the target project root.

After copying, the target project should contain these files and directories at its root:

AGENTS.md
.agents/
.codex/
assets/
docs/
examples/
MANIFEST.md
README.md
README_CN.md

This full-pack install may overwrite an existing AGENTS.md, README.md, or .codex/config.toml. Back up or merge those files first if the target project already uses them.

Then open a new Codex session from that project root so Codex can load the project rules, Skills, and Agents.

macOS / Linux command line, optional

Most users do not need this. Use it only if you prefer Terminal and already have the pack downloaded locally.

From inside the extracted pack directory:

TARGET="/absolute/path/to/target-project"

mkdir -p "$TARGET"
rsync -av --exclude ".git/" ./ "$TARGET"/

Replace TARGET with the real target project path. This copies the complete pack, including runtime files, docs, examples, assets, readmes, and manifest.

Runtime-only install

Use this when the target project already has its own README/docs and you only want Codex to load the founder workflow runtime:

PACK="/absolute/path/to/ai-native-founder-skill-pack"

mkdir -p .agents/skills .agents/shared .codex/agents

cp -R "$PACK/.agents/skills/." .agents/skills/
cp -R "$PACK/.agents/shared/." .agents/shared/
cp -R "$PACK/.codex/agents/." .codex/agents/
cp "$PACK/.codex/config.toml" .codex/config.toml
cp "$PACK/AGENTS.md" AGENTS.md

The /. suffix is intentional: it copies the contents of each directory and avoids creating nested paths such as .agents/skills/skills.

Open a new Codex session from the target project after copying so the project rules, Skills, and Agents are visible.

Verify

Ask Codex:

List the AI Native Founder skills and custom agents you can see. Then summarize
the AGENTS.md rules you loaded.

Run the deterministic helper scripts:

python .agents/shared/scripts/stage_gate_check.py examples/sample-evidence.json
python .agents/shared/scripts/pmf_signal_score.py examples/sample-pmf-metrics.json

Expected result: both commands print structured output and a short summary.

Documentation

• docs/agent-skill-map.md explains how agents and skills fit together.
• docs/design-principles.md documents the operating principles behind the pack.
• docs/installation.md covers installation shapes and runtime details.
• docs/promptbase-packaging.md covers marketplace packaging guidance.
• docs/usage-examples.md provides example prompts for common founder workflows.
• examples/ contains sample stage outputs and JSON inputs for the helper scripts.