Skip to content
View samuelrojas-dev's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report samuelrojas-dev

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
samuelrojas-dev/README.md

Hello, I'm Samuel

Developer focused on AppSec & AI Agent security

I build open source security tools — from vulnerability triage to adversarial testing for LLM agents. I design and run benchmarks against prompt injection, data leakage, MCP server exploits, and cost-based attacks. I also ship production apps with Flutter and Python. Currently building depquiet, an SCA triage tool that cuts through vulnerability alert noise.

Projects

  • depquiet · PyPI · codecov: an SCA triage tool that cuts through vulnerability alert noise. 6 parsers (npm, pip, trivy, grype, dependabot, snyk), EPSS scoring, AST-based reachability analysis, 21 community templates, SARIF output, GitHub Action — 292 tests, 100% coverage, published on PyPI.
  • rogue-llm · PyPI: an adversarial testing toolkit for LLM agents. Ships a hostile model server, 7 attack profiles, a security scanner, MCP server auditing, a framework benchmark with badge generation, and a pytest plugin — all offline, no API keys needed. 114 tests, published on PyPI.
  • dowbench: a reproducible benchmark of denial-of-wallet attacks against tool-using LLM agents. Found a real 8.8× cost-amplification attack against Gemini and the one defense (out of five tested) that stops it — every result is replay-verified and reproducible from a published cassette, no API key needed.
  • agent-security-lab: a B2B sales agent built on Gemini and Supabase, evaluated against prompt-injection and information-disclosure attacks, with 34 adversarial test cases and an automated evaluator.
  • Checho: an offline-first financial education app built with Flutter, with a CI pipeline on GitHub Actions.

What I'm working on

  • Building depquiet: vulnerability alert triage for DevSecOps teams — filtering noise from SCA scanners so developers fix what matters
  • Building rogue-llm: adversarial testing for LLM agents, MCP servers, and agentic frameworks
  • Building with Flutter, Supabase, and Python

Tools

Python Flutter Dart Supabase SQL Burp Suite GitHub Actions Git

Certifications and languages

  • Google Cybersecurity Certificate
  • English (C1, EF SET) and Spanish

Build it well, then make it hard to break.

Contact

LinkedIn

Pinned Loading

  1. depquiet depquiet Public

    Tells you what to change to close your dependency alerts, verified with your resolver, and which alerts never reach production.

    Python 3

  2. depquiet-fixpr-test depquiet-fixpr-test Public

    Demo: depquiet opens one verified PR that fixes 10 alerts Dependabot can't (botocore holds urllib3 back). uv and Poetry.

    Python

  3. agent-cost-bench agent-cost-bench Public

    Reproducible denial-of-wallet (DoW) benchmark for tool-using LLM agents, scored on provider-billed tokens/USD.

    Python 2 1

  4. agent-security-lab agent-security-lab Public

    Attacking an LLM sales agent with 23 prompt-injection attacks: no defense vs. prompt-only vs. data-layer isolation

    Python 1

  5. rogue-llm rogue-llm Public

    Adversarial LLM proxy for testing structural vs behavioral agent security. Replaces the model with a hostile one — 7 attack profiles, MCP server testing, security benchmark.

    Python 1