Open-source PuedoBot uses an adversarial second LLM pass to triage SAST false positives
No-Bumblebee-3536 · reddit · 2026-10-17
The author open-sourced PuedoBot (Apache 2.0), a CLI security auditor that cuts SAST false positives with an adversarial two-pass LLM pipeline.
Pipeline design:
- Deterministic sweep: AST sink traversal, Bandit, Semgrep rules, OSV.dev dependency checks, and git-history secret scanning (auto-redacted).
- Quarantine: only flagged snippets are sent to the LLM, wrapped as untrusted input to block prompt injection from malicious comments or strings — no whole-repo dumps.
- Pass 1 (The Scout): traces data from source to sink, checks for missing sanitizers, estimates blast radius.
- Pass 2 (The Defense Attorney): a separate pass whose sole job is to refute Pass 1 by finding framework-level protections, upstream guards, or dead code paths. Only findings that survive the argument reach the user.
Other details:
- Model-agnostic: works with OpenAI, DeepSeek, Gemini, Anthropic, Groq, Mistral, OpenRouter, or local models via Ollama/vLLM/LM Studio.
- Extracted from a larger private CLI that adds autonomous repo discovery (6-month cooldown per repo), model failover, and responsible disclosure — kept private for safety reasons.
- Stack: Python 3.11+, FastAPI/uvicorn, PostgreSQL, Bandit, Semgrep (optional), OSV.dev.
More from coding & agent
- Screenshot to spreadsheet in 11 seconds with one line of PowerShell — dfinke · 2026-10-17
- Deploy MiniCPM-V 4.7 locally by pointing Claude at its Hugging Face page — NielsRogge · 2026-10-17
- ThePrimeagen mocks AI coding: infinite tokens still produce an Electron app — seatedro · 2026-10-17
- Dev uses Unreal Engine MCP server with an agent to build a virtual workplace in an hour — Teknium · 2026-10-17
- Generative UI 101: Building Live Interfaces with Agents on Opus 5.5 and Rivals — EXM7777 · 2026-10-17
- MCP server review: gate tool writes on auth/traversal findings, not test score — SamuelFmedeiros · 2026-10-17