Pentest Agent Suite Bug Exposes Claude Code and 6 AI Coding Tools

Key Takeaways

• A new open-source, autonomous bug bounty framework, Pentest Agent Suite, has been released on GitHub.
• The suite integrates with seven major AI coding platforms, including Claude Code, OpenAI Codex, and Google Gemini.
• It features 50 specialized security agents, 26 slash commands, and 19 CLI tools, designed to automate vulnerability discovery and reporting.
• The framework emphasizes rigorous validation processes for findings and offers extensive configuration options for penetration testing.

A significant development in the realm of automated cybersecurity testing has emerged with the open-sourcing of the Pentest Agent Suite, a comprehensive bug bounty framework. This new platform, designed for fully autonomous operation, integrates with a broad spectrum of AI coding environments and aims to streamline the vulnerability discovery and reporting process for security researchers.

Published on GitHub by researcher H-mmer, the suite is a robust security platform equipped with persistent memory and real-time integration with leading bug bounty platforms. It also features a FAISS-backed semantic writeup search engine, allowing agents to query existing vulnerability knowledge before initiating new tests.

The Pentest Agent Suite is structured into three core components: a collection of 50 specialized agents, a dual-server Model Context Protocol (MCP) infrastructure, and an extensive library of rules and methodologies.

Its bounty-platforms MCP server facilitates interaction with 16 programs, including prominent platforms like HackerOne, Bugcrowd, Intigriti, Immunefi, and YesWeHack. This integration provides access to seven key tools, such as list_platforms, get_program_scope, sync_program, draft_report, and submit_report, enabling seamless program management and report submission.

The writeup-search MCP server intelligently supports three modes for retrieving information: FAISS semantic search, SQLite keyword search, and a local fallback mechanism that queries the bundled rules/payloads.md. This file contains 2,605 lines detailing various attack patterns, including XSS, SSRF, SQLi, IDOR, OAuth, SSTI, JWT, LFI, prototype pollution, NoSQLi, and DeFi vulnerabilities.

Pentest Agent Suite Framework

A standout feature of the framework is its “7-Question Gate,” a stringent validation pipeline managed by the validator agent. Every potential finding must pass this gate; a single “NO” response automatically triggers a verdict of KILL, DOWNGRADE, or CHAIN REQUIRED, preventing unverified issues from proceeding.

Crucially, no finding can be submitted without a /validate PASS status and a /quality score of 7 or higher. These criteria are enforced by hard gates within the /report and /submit commands, ensuring a high standard of reported vulnerabilities.

The /autopilot command incorporates an anti-shallow depth engine, mandating multi-layer stacked encoding for every payload attempt. It refuses to declare an attack surface exhausted until a complete exhaustion matrix is achieved, configurable via --paranoid, --normal, or --yolo checkpoint modes.

A persistent brain.py module tracks every endpoint per target, implements circuit-breaker logic (triggering a 60-second auto-backoff after five consecutive 403/429 responses), and synchronizes knowledge across engagements using incremental hash-based diffing.

The suite’s installer, invoked via python3 -m tools.installer, automatically generates native configuration formats for each supported AI coding tool and places them in the appropriate IDE directories.

For IDEs lacking native subagent support, specifically Cursor, Windsurf, and OpenClaw, the framework translates content into skill files and rules. This process involves stripping Claude-specific prose and rewriting path variables to absolute references, ensuring compatibility.

50 Agents Across Five Tracks

The framework boasts a diverse roster of agents, categorized across various security tracks. This includes 19 HackerOne weakness specialists, such as xss-hunter, sqli-hunter, ssrf-hunter, rce-hunter, oauth-hunter, and llm-ai-hunter. An 8-agent SAST pipeline is also included, alongside infrastructure and reconnaissance agents like cloud-recon, js-analyzer, graphql-audit, and waf-profiler. Additionally, a web3-auditor is dedicated to identifying Solidity and DeFi patterns.

Five deep methodology skills complement these hunters, each derived from hundreds of successful paid reports. These include hunt-rce (covering vulnerabilities like RSC CVE-2025-55182, runc Leaky Vessels, and BentoML pickle deserialization), hunt-xss (addressing DOMPurify mXSS and n8n MCP OAuth XSS GHSA-537j-gqpc-p7fq), and hunt-llm-ai, which aligns with OWASP LLM Top 10 v2025 and the Agentic AI Top 10.

Cost tracking is managed through CC hooks, with the SubagentStop event triggering cost_hook.py to log agent names and session costs to cost-tracking.json, providing live spend visibility in the status line.

A PreToolUse scope hook (scope_hook.py) matches every Bash command against scope.yaml using exact and wildcard patterns, preventing out-of-scope execution before any tool call. CVSS scoring is programmatically enforced by cvss_version_guard.py, which mandates CVSS 3.1 for HackerOne and CVSS 4.0 for all other platforms.

Quick Start

To begin using the Pentest Agent Suite, users can follow these steps:

export HACKERONE_USERNAME=you HACKERONE_TOKEN=your_token
uv run python3 tools/scaffold.py hackerone tesla
cd ~/bounties/hackerone-tesla && claude
/hunt tesla.com

The framework requires Python 3.10+ and uv, along with standard reconnaissance tools such as nmap, httpx, subfinder, nuclei, ffuf, katana, and sqlmap.

The framework is available on GitHub and is licensed exclusively for authorized security testing under responsible disclosure. A bundled rag-builder/ utility allows for the construction of local FAISS writeup indexes from a seed list of 146 repositories, including CTF archives, bug bounty reports, and payload collections. All destructive operations within the suite are gated behind an explicit --execute flag.

What You Should Do

• Review the Pentest Agent Suite’s documentation thoroughly before deployment to understand its capabilities and limitations.
• Ensure all necessary dependencies, including Python 3.10+ and the specified recon tooling, are installed and properly configured.
• Adhere strictly to the licensing terms, which permit use only for authorized security testing under responsible disclosure.
• Utilize the --execute flag cautiously for any destructive operations, fully understanding their implications.
• Regularly update the framework and its components to benefit from the latest security enhancements and bug fixes.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.