Skip to content
View xeevees-lab's full-sized avatar
:electron:
React is reacting...
:electron:
React is reacting...

Highlights

  • Pro

Block or report xeevees-lab

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
xeevees-lab/README.md

◆  GARDIAN  ◆

Graph-Augmented RAG Detection with Intelligent Agent Network

The problem

Static analysers are loud. Semgrep will happily hand you two hundred findings, most of which aren't reachable, and a reviewer stops reading at forty.

What GARDIAN does

It doesn't detect harder — it adjudicates. Every candidate gets checked for real reachability, grounded against CWE/OWASP evidence, argued over by opposing agents, and if it survives all that, patched with a fix that has already been executed in a sandbox to prove it works.

Semgrep CE runs as the baseline arm, so the improvement is measured, not asserted.

The pipeline

  merge request
       │
       ▼
  ① Semgrep CE ........ candidates
       │
       ▼
  ② Neo4j + Cypher .... taint reachable?
       │
       ▼
  ③ RAG ............... CWE · OWASP · CVEfixes
       │
       ▼
  ④ LangGraph ......... agents adjudicate
       │
       ▼
  ⑤ Docker sandbox .... patch verified
       │
       ▼
  GitLab CI posts the fix
 Design constraints, and why they matter 
Constraint Consequence
Zero cash No paid APIs, no cloud GPUs, no managed services. Every component is free-tier or open source.
Self-hosted Neo4j, the sandbox and the CI runner all live in local Docker on WSL2. Nothing leaves the machine.
One semester Aug–Nov 2026, vertical slice first — walking skeleton before breadth.
Air-gappable A security tool you can't run on your own infrastructure is a security tool most teams can't adopt.

The constraints aren't limitations I worked around — they're the reason the thing is adoptable. A PR auditor that ships your source code to a third-party API is a non-starter for exactly the teams that need one most.


Python · Pydantic v2 · Semgrep CE · tree-sitter · Neo4j · Cypher · LangGraph · sentence-transformers · Docker · GitLab CI

◆  Stack  ◆

Infra & CI
Security & Data
Languages

◆  Activity  ◆

Most of my day-to-day commits live on GitLab — the numbers below only tell part of the story.






contribution snake

"If it works on my machine, it works on yours."



Pinned Loading

  1. pramana pramana Public

    pramāṇa (प्रमाण) — "a valid means of knowledge." A news intelligence platform that separates verified fact from contested claim, shows how outlets frame the same event differently, and grounds ever…

    JavaScript

  2. network-intelligence-dashboard network-intelligence-dashboard Public

    FastAPI tool for single-host reconnaissance: threaded TCP port scan over a configurable range, IP geolocation, persistent scan history with aggregate stats and CSV export. Vanilla-JS frontend, no n…

    HTML