T3MP3ST: Turn Claude Code & Codex into Autonomous Red-Team Hacker Swarms (Open Source)

T3MP3ST is an open-source framework that transforms existing AI coding agents into autonomous red-team swarms. Released by Pliny the Liberator under AGPL-3.0, it automates the full kill chain from reconnaissance to exploitation and reporting within a single War Room interface. Public benchmarks on CTF and CVE tasks demonstrate its practical value for authorized security research.

📑Table of Contents

1. T3MP3ST Overview and Background
2. Kill Chain and 8-Operator Architecture
3. Benchmark Results and Reproducibility
4. Installation and War Room Usage
5. Supported Domains and Current Status
6. Usage Caveats and Authorization Requirements
7. Frequently Asked Questions (FAQ)
8. Summary and Recommended Next Actions

---

T3MP3ST Overview and Background

T3MP3ST repurposes agents such as Claude Code and Codex as “hacker swarms” without requiring new API keys or cloud spend in keyless mode. The project emphasizes zero-day hunting accessibility and publishes all source code and verification artifacts on GitHub.

Key characteristics include:

• Reuse of existing user agents (no additional credentials needed)
• End-to-end kill-chain automation
• Real-time War Room UI for mission control
• Evidence Vault with OPSEC controls

According to the official README, the framework also supports local models via Ollama, enabling cost-effective large-scale testing. The announcement on X highlighted strong results on XBOW and Cybench suites.

---

Kill Chain and 8-Operator Architecture

The core design consists of eight specialized operators that collaborate through an Evidence Vault and OPSEC layer.

Operator	Role	Status
Recon	Target discovery and information gathering	Stable
Scanner	Vulnerability scanning	Stable
Exploiter	Exploit execution	Experimental
Infiltrator	Intrusion and privilege escalation	Experimental
Exfiltrator	Data exfiltration	Experimental
Ghost	Trace concealment	Planned
Coordinator	Task orchestration	Stable
Analyst	Report generation	Stable

The table summarizes the status matrix from the repository. While Recon and Coordinator modules are production-ready, later-stage operators remain experimental. Source: https://github.com/elder-plinius/T3MP3ST (as of July 2026)

---

Benchmark Results and Reproducibility

The project publishes the following reproducible results:

• XBEN (104-challenge suite): 90.1% pass@1 mean (Wilson-95 CI 86.2–92.9)
• Cybench (40 tasks, hint-free): 23/40 (58%) with Opus 4.8
• CVE-Zero (10 post-cutoff CVEs across 7 languages): 8/10 exact file/line/CWE match for single-agent mode

All headline numbers can be re-derived with npm run verify-claims. The repository contains committed JSON artifacts that independently verify all 24 claims.

---

Installation and War Room Usage

Setup is straightforward:

1. Clone the repository
2. Run npm install && npm run server
3. Open http://127.0.0.1:3333/ui/ in a browser

Once a local agent is connected to the War Room, users can create missions, monitor execution, and generate reports through the GUI. Keyless mode allows fully local operation with Ollama or similar backends.

---

Supported Domains and Current Status

T3MP3ST targets Web, CTF, CVE, Cloud, DeFi, Binary, and Mobile environments. Stable components currently include the Recon engine, Mission engine, Arsenal (35+ tools), Egress-scope containment, and Coordinated-disclosure pipeline. Cloud, Mobile, and Binary modules are either experimental or planned.

---

Usage Caveats and Authorization Requirements

Because T3MP3ST is an offensive security tool, it must only be used against systems with explicit authorization. The AGPL-3.0 license does not grant permission for unauthorized testing. Security teams should evaluate internal policy alignment before adoption.

---

Frequently Asked Questions (FAQ)

Q: Can T3MP3ST be used with agents other than Claude Code?

Yes. It supports Codex, Hermes, and local models through Ollama or vLLM. Any agent that can connect to the War Room interface is compatible.

Q: Are the benchmark numbers truly reproducible?

Yes. Running npm run verify-claims recomputes every metric from the committed JSON artifacts. All 24 claims have been independently verified.

Q: Is it safe to use against unauthorized targets?

No. The documentation explicitly requires “explicit permission.” Unauthorized use carries legal and ethical risks and should be avoided.

---

Summary and Recommended Next Actions

T3MP3ST offers a practical way to leverage existing AI agents for red-team automation. Its reproducible benchmarks and convenient War Room UI are clear strengths, yet its offensive nature demands strict authorization controls.

Recommended next steps for readers:

• Clone https://github.com/elder-plinius/T3MP3ST and test locally
• Run npm run verify-claims to reproduce the published benchmarks
• Review internal security policies for alignment
• Compare with similar frameworks such as Arbor

Related articles:

• Arbor: Hypothesis-Tree AI Optimization Framework Beats Claude Code & Codex by 2.5x [2026]
• Google Agent Development Kit (ADK) Open Source Release — Production-Grade Multi-Agent Framework
• Meta Restricts Claude Code and Codex for Applied AI Engineers to Prevent Distillation into Llama Models

Related new article:

• Codex Plugin for Claude Code Review and Rescue Setup – This published update adds current operational context for T3MP3ST: Turn Claude Code & Codex into Autonomous Red-Team Hacker Swarms (Open Source).
• Keystroke n8n Alternative for Claude Code and Cursor: Setup and Fit Check – This published update adds current operational context for T3MP3ST: Turn Claude Code & Codex into Autonomous Red-Team Hacker Swarms (Open Source).
• Docker Sandboxes (sbx) Guide: Run Claude Code Safely in a microVM – This published update adds current operational context for T3MP3ST: Turn Claude Code & Codex into Autonomous Red-Team Hacker Swarms (Open Source).