Local intentionally vulnerable lab with a guided workshop and CTF challenges for practicing Git push-option RCE, unsafe deserialization, duplicate-key parsing, and configuration overrides against a Dockerized Git server.

A guided workshop that teaches push options, unsafe serialization, duplicate-key parsing, configuration override, and the constrained execution chain. Lessons combine short explanations with safe experiments against the running lab.

Research Lab

A CTF-style experience with three sequential Challenge cards, optional hints, and backend-generated flags. The normal UI does not expose source, traces, serializer output, execution-gate output, or a guided walkthrough.

Quick start

No flag or credential setup is required. A one-shot initializer creates independent random proof values and service credentials in a private Docker volume on first startup. make up is an equivalent convenience command. An optional .env is needed only to override defaults or keep explicit values outside Docker.

The disposable lab account is:

These public credentials exist only inside the local, disposable lab. Do not reuse them elsewhere.

Requirements

• Docker Engine or Docker Desktop
• Docker Compose v2 (docker compose)
• Git
• Approximately 4 CPU cores, 6 GB of memory available to Docker, and 8 GB of free disk space for the first build

The release-readiness pass is verified on Apple Silicon (arm64). Other operating systems and CPU architectures are not claimed as verified until their clean-clone checks are recorded.

Basic operation

make reset deletes Docker volumes, including lab repositories, runtime progress, proof values, and internal credentials. The next startup generates fresh private values automatically.

Safety model

Published ports bind to 127.0.0.1. Redis, the hook executor, and internal processing services are not published. The executor has no Docker socket or host filesystem mount, drops Linux capabilities, uses a read-only filesystem, accepts no shell command or caller-controlled environment, and can invoke only exact lab-owned proof executables. These controls constrain the exercise; they do not make it safe to expose publicly.

Architecture

See the system overview, push-option processing, threat model, and local deployment guide.

Scope disclaimer

ForgeDock is a deliberately simplified and safely constrained reconstruction of the publicly documented vulnerability concepts behind CVE-2026-3854. It does not reproduce GitHub’s proprietary implementation. Field names, service boundaries, and execution behavior differ from the actual GitHub and GitHub Enterprise Server implementations.

The intentionally vulnerable behavior was added downstream for this lab and is not an upstream Forgejo vulnerability. ForgeDock is not GitHub, and this project is not affiliated with or endorsed by GitHub, Forgejo, Wiz, or the cited researchers.

Credits

• Sagi Tzadik and Wiz Research for the public vulnerability research that inspired the learning concepts.
• Forgejo and its contributors for the downstream server base. Exact provenance is in docs/upstream-forgejo.md.
• The contributors to FastAPI, Redis, and ttyd, used by the local packaging.

These credits describe sources and dependencies; they do not imply collaboration, approval, or endorsement.

License and notices

Flagship-created software and curriculum are licensed under GPL-3.0-or-later, except where a file states another compatible license. See LICENSE, NOTICE.md, Forgejo provenance, and TRADEMARKS.md. Existing third-party licenses and notices remain in force.