Vibe Coding DiscoverTORCH
View on GitHubKarpathy LLM based claude harness for PenetrationTesting / Bugbounty using obsidian
TORCH is a Claude Code harness that turns an Obsidian vault into a pentest/bug-bounty second brain: 500+ wiki pages, per-vulnerability hunt skills, deterministic hooks, an MCP semantic-search layer, and scripts/campaign.py driving whole engagements autonomously.
Use Cases
Autonomous end-to-end penetration test campaignsBug bounty program recon and huntingCTF / boot-to-root box solvingPer-vulnerability hunt skills (XSS, SQLi, SSRF, IDOR, RCE, deserialization)Active Directory and cloud (AWS/Azure/GCP) attack pathsLLM/AI, API, ICS/OT and request-smuggling testingCVE discovery on binaries, libraries and reposSemantic search over a 500+ page offensive-security wikiEngagement state tracking and kill-chain coverage boardsScope guardrails and rules-of-engagement enforcementPoC evidence capture and lintingClient report generation for pentestsClient-data leak gating before push
Built With
- Language
- Python
- Frameworks
- Claude Code · Model Context Protocol · Obsidian · qmd · Codex-style skills · pytest · Bun · Node.js
Tags
claude-code · claude-skills · pentesting · bug-bounty · offensive-security · red-team · ctf · mcp · obsidian · knowledge-base · semantic-search · agent-harness · hooks · security-automation