5 security tools shipped. Each one solves a real problem.
Autonomous HTB/CTF pentest agent: recon → enumeration → web → exploit → privesc → report. Tiered Claude with prompt caching. NSE vuln + hydra brute + SSH-based privesc. Human-in-the-loop checkpoints or full auto mode.
Autonomous vulnerability discovery pipeline: AFL++ fuzzing → ASan/GDB triage → LLM root cause analysis → patch generation → validation. Inspired by DARPA AIxCC.
Deduplicates Semgrep/Bandit/ZAP findings by CWE+location, scores risk 0-10, runs a local LLM to filter false positives, and exports SARIF 2.1.0 for GitHub Code Scanning.
AI-assisted STRIDE threat modelling tool that generates executive-quality HTML reports with MITRE ATT&CK mapping, compliance flagging (GDPR/PCI-DSS), and on-prem LLM narratives from YAML, OpenAPI, Dockerfile, or docker-compose inputs.
Parses Linux auth logs and journalctl JSON, detects 7 attack patterns mapped to MITRE ATT&CK with confidence scoring, enriches IPs via AbuseIPDB, and sends a local LLM threat assessment.
HackTheBox and TryHackMe machine writeups with custom exploit scripts. Commit-on-root discipline: every rooted machine gets a writeup pushed to GitHub.
Hands-on C exploitation exercises — stack buffer overflows, heap corruption, use-after-free. GDB + pwndbg for dynamic analysis. Foundation for OSCP exploit development.