🧪 Token Launch Lab

An AI red team that tries to kill your token launch — before mainnet does.

Token launches are irreversible. One bad vesting schedule, one missed exploit vector, one weak narrative, or one regulatory blind spot can torch a launch in public. Token Launch Lab runs adversarial AI agents that attack your launch from every angle, then a Judge Agent verifies every finding before it ships.

   ┌─────────────┐   ┌──────────────┐   ┌────────────────┐   ┌──────────────┐
   │  DUMP RISK  │   │ PROTOCOL RISK│   │ REGULATORY RISK│   │  CT ADVERSARY│
   └──────┬──────┘   └──────┬───────┘   └───────┬────────┘   └──────┬───────┘
          └─────────────────┴─────────┬─────────┴───────────────────┘
                                       ▼
                            ┌──────────────────────┐
                            │   ⚖️  JUDGE AGENT      │
                            │  verify · score · kill│
                            └──────────┬───────────┘
                                       ▼
                              📄  KILL REPORT

---

💡 The Thesis

The future crypto founder does not need more AI that agrees with her. She needs an AI that tries to kill the launch first — so reality doesn't get the chance.

Most AI tools help you build launch materials. Token Launch Lab does the opposite: it's a Codex-backed adversarial agent harness, not another friendly chatbot. Codex runs specialized agents that each try to break your launch from a different direction, write inspectable markdown reports, and submit them to a local Judge Agent that rejects anything without traceable evidence.

---

⚔️ Meet the Red Team

Agent	Attacks	Writes
📉 Dump Risk	Investor/team unlocks, TGE float, liquidity & MM allocation, dump pressure	outputs/dump-risk.md
🛡️ Protocol Risk	Audit status, pause policy, multisig, incident response readiness	outputs/protocol-risk.md
⚖️ Regulatory Risk	Public-sale flags, jurisdiction ambiguity, incentive-campaign risk	outputs/regulatory-risk.md
🐦 CT Adversary	Narrative fragility, likely crypto-Twitter criticism, token-necessity & farming optics	outputs/ct-adversary.md
🧑‍⚖️ Judge / Orchestrator	Verifies schema, evidence, severity, safety boundaries; computes readiness	outputs/kill-report.md

Every finding must cite an exact quote from tge-spec.md. No evidence → the Judge rejects it.

---

🚀 60-Second Quickstart

npm install
node src/orchestrator.js   # run the Judge Agent over the agent reports
npm run ui                 # launch the live demo UI

The orchestrator reads the Codex-authored reports in outputs/, verifies them, prints pass/revision status, computes launch readiness, and writes:

• 📄 outputs/kill-report.md — ranked failure modes + readiness score
• 🔧 outputs/remediation.md — prioritized defensive fixes
• 🧑‍⚖️ outputs/judge-evaluation.md — the verification verdict

Then open the dashboard:

http://127.0.0.1:3000

---

🏗️ Architecture

Codex /goal
  ├─ reads   codex-goal.md · AGENTS.md · tge-spec.md
  ├─ writes  outputs/dump-risk.md
  ├─ writes  outputs/protocol-risk.md
  ├─ writes  outputs/regulatory-risk.md
  ├─ writes  outputs/ct-adversary.md
  └─ Judge Agent → node src/orchestrator.js
                     └─ writes kill-report.md · remediation.md · judge-evaluation.md

Why this is a real harness, not a prompt:

• 🎯 Codex integration — codex-goal.md is the exact /goal prompt.
• 🤝 Multi-agent orchestration — AGENTS.md defines five role contracts.
• 🧠 Inspectable markdown memory — agents write to outputs/, nothing hidden.
• ✅ Verification — src/orchestrator.js checks fields, evidence, severity, confidence, safety.
• 🏁 Termination — the run completes only when all reports pass.
• ♻️ Recovery loop — failed reports are revised individually and re-checked.

---

📊 Scoring

Dimension	Values
Severity	low · medium · high · critical
Confidence	low · medium · high
Evidence	every finding cites tge-spec.md
Remediation priority	P0 · P1 · P2
Launch readiness	0 – 100

---

🎬 Live Demo Flow

1. Open http://127.0.0.1:3000.
2. Show the HarborUSD Kill Report and the launch-readiness score.
3. Click Run Judge Verification to run node src/orchestrator.js from the UI.
4. Show codex-goal.md, AGENTS.md, and tge-spec.md inside the evidence panels.
5. Open outputs/kill-report.md if judges want the raw markdown.

---

🛡️ Safety Boundaries

This is defensive launch-risk review only.

• ❌ Does not generate exploit instructions.
• ❌ Does not provide legal advice — output is risk flags for qualified review.
• ❌ Does not produce harassment, misinformation, or market-manipulation content.
• ✅ All findings are written to inspectable markdown — nothing hidden in opaque state.

---

🧰 Legacy Canned Demo

The older vesting-calculator artifact demo is still available:

npm run build
npm run demo
npm run test:artifact
node ./dist/index.js dashboard -p 3000

---

📂 Repo Map

Path	What's inside
codex-goal.md	The exact Codex /goal prompt
AGENTS.md	Agent role contracts + finding schema
tge-spec.md	Shared input the agents red-team
src/orchestrator.js	The Judge Agent / verifier
outputs/	Inspectable agent memory + final reports
web/	Live demo dashboard
PITCH.md · SUBMISSION.md	The story & the submission

---

Built for crypto founders who'd rather be killed in private than in public.

Read the full pitch → PITCH.md

Contributors: ElenaX <leini8891@qq.com>