Skip to content
View Judge-M's full-sized avatar
5 followers · 11 following

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex2Achievement: YOLO

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex2Achievement: YOLO

Block or report Judge-M

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. Markdown is supported. 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
Judge-M/README.md 
      ██╗██╗   ██╗██████╗  ██████╗ ███████╗    ███╗   ███╗
      ██║██║   ██║██╔══██╗██╔════╝ ██╔════╝    ████╗ ████║
      ██║██║   ██║██║  ██║██║  ███╗█████╗      ██╔████╔██║
 ██   ██║██║   ██║██║  ██║██║   ██║██╔══╝      ██║╚██╔╝██║
 ╚█████╔╝╚██████╔╝██████╔╝╚██████╔╝███████╗    ██║ ╚═╝ ██║
  ╚════╝  ╚═════╝ ╚═════╝  ╚═════╝ ╚══════╝    ╚═╝     ╚═╝


$ run judge.sys
> load core.module ............... 100%
> mount /docket .................. 16 repos
> verify chain ................... 1,704 commits · signed
> uplink ......................... established

INCOMING TRANSMISSION — MATTHEW J.
builds the infrastructure AI agents actually run on:
memory they can trust · tools they must earn · compute they can schedule

signal followers linkedin

> whoami

Hey, I’m Matt Judge 👋

I build open-source infrastructure for AI agents, with a focus on trusted memory, scoped permissions, and systems people can inspect and control.

My projects explore how agents remember, use tools, coordinate work, and run on local compute. That includes BrainConnect, AgentConnect, and Decima, built around a belief that useful autonomy needs clear boundaries and accountable decisions.

I work across Python, Rust, and TypeScript, with interests spanning local AI, cybersecurity, and hands-on learning. Alongside agent infrastructure, I build CTF challenges, incident-response exercises, and tools that help people learn to work with AI.

I learn by building, testing, and sharing what I discover. This GitHub is where that work takes shape.

> docket --featured

001 · Decima ACTIVE An agent-native operating system — four verbs over one signed append-only log, where state is a fold and authority is a held object. Python event-sourcing capability-security · 625 commits · 1,262 tests + Playwright e2e

002 · AgentConnect RC Control plane for managed coding agents — tasks, artifacts, reviews, audits, handoffs. Verified driving a real claude agent end-to-end; an auth-bypass bug found, fixed, retested. Python Apache-2.0 · 9 packages · ~1,500 tests

003 · BrainConnect RC Human-gated trusted memory for agent systems — agents propose, humans promote. Scoped claims, provenance, contradiction, supersession. Retrieval can never widen trust. Python MCP Apache-2.0 · 951 tests · ★ most-starred on the docket

004 · CTFGenerator ACTIVE AI-resistant CTF platform — deterministic, seed-based challenges grounded in real CVEs, behind a full FastAPI + PostgreSQL control plane with 8-role RBAC and isolated build workers. Python FastAPI PostgreSQL · 270 commits

005 · Operation Silent Ridge LIVE A complete cooperative incident-response event for 10 teams — Autopsy, Wireshark, Cutter, Wazuh, IRIS + CTFd, browser desktops via Guacamole, all Dockerized and drill-tested. Currently shipping for a live late-October event.

> docket --archive · supporting planes & side missions
  • Seam — minimal Rust runtime: orchestrator delegates to sandboxed workers, all authority through a capability broker
  • Connect-Governance — deterministic decision kernel: pure evaluate(), byte-replayable records, Ed25519 grants
  • ToolConnect — tool-governance decision point: Cedar policies, tamper-evident audit, MCP discovery
  • ComputeConnect — local compute orchestration with default-deny privacy tiers over llama.cpp engines
  • Connect-Control — server-rendered control-plane app over the Connect ecosystem
  • Connect — ecosystem hub: architecture, ADRs, engineering standards
  • OmniDM — AI dungeon master with a real rules engine; web, desktop, mobile, CLI + Discord/Slack/Matrix bots · MIT
  • brainconnect-control-model — LoRA SFT on Qwen3-4B, adversarial-hardened, measured-lossless Q4_K_M CPU quant
  • TheJudgeLadder — learn to work with AI from zero: five rungs, first chat to supervising agents
  • CEI-Labs-Agent — beginner-friendly local CTF coach, Ollama-powered, teaches rather than solves

> evidence

GitHub stats Contribution streak

> arsenal

languages   Python · Rust · TypeScript · Shell
agents      MCP · event-sourcing · capability security · audit chains
inference   Ollama · llama.cpp · LoRA SFT · Q4_K_M quantization
platform    Docker · FastAPI · PostgreSQL · SQLite · GitHub Actions
security    Wazuh · IRIS · CTFd · OIDC/PKCE · secret & injection scanning

> tip

If something on the docket helped you, fuel the next build:

GitHub Sponsors Buy Me a Coffee

> contact

LinkedIn GitHub

> end of transmission
state is a fold · authority is held · judgment is human

Popular repositories Loading

  1. BrainConnect BrainConnect Public

    A human-gated trusted memory ledger for agent systems, with scoped claims, provenance, contradictions, and supersession.

    Python 3

  2. Decima Decima Public

    An agent-native operating system: four verbs over one append-only log, where state is a fold and authority is a held object.

    Python 1

  3. CEI-Labs-Agent CEI-Labs-Agent Public

    Python 1 1

  4. Seam Seam Public

    A minimal runtime for building agent systems from clean, replaceable seams instead of one giant harness.

    Rust 1

  5. Oct26 Oct26 Public

    Python 1 1

  6. OmniDM OmniDM Public

    TypeScript