galyarder-security
galyarder-security (formerly Shannon Proxy) is a Docker/Temporal-based auditing layer that intercepts model SDK calls (Anthropic, Groq, local Ollama, etc.) and routes them through a local proxy on port 8317 rather than direct cloud endpoints, giving operators visibility into and control over what autonomous coding/tool-using agents execute. It runs structured "security sweep" and threat-modeling workflows against a target repo (`./shannon start`) and includes deterministic cleanup routines (`./shannon cleanup`) to remove agent-generated scripts and execution residue and prevent workspace pollution/exposure. It targets a threat model broader than chat-based prompt injection: agent tool-execution drift, dependency supply-chain issues, and unverified code paths in "agentic company" shell-access workflows.
What it is
galyarder-security (formerly Shannon Proxy) is a Docker-based local proxy plus Temporal workflow engine that sits between an autonomous agent runtime and model APIs. The summary covers its purpose; the page adds the concrete deployment shape, prerequisites, and operating commands an engineer needs before trying it.
How it works
It runs a local auditing proxy on port 8317. The agent/worker runtime points its model SDK at that proxy instead of cloud endpoints; the proxy then fans out to Anthropic, Groq, or local Ollama. "Shannon CLI Proxy patches hardcoded model SDK constraints at build time" — the model identifier is injected via a Docker build arg (DEFAULT_MODEL), and endpoint/auth come from .env (ANTHROPIC_BASE_URL, ANTHROPIC_API_KEY). Security sweeps and cleanup run as Temporal workflows that store "verifiable evidence of security state."
Preconditions / Affected
You must already have Docker + Docker Compose and a separate local LLM proxy (CLIProxyAPI, LiteLLM, or OpenRouter) with a model configured — this tool does not itself talk to cloud model APIs directly; it assumes your proxy does. It only matters if your agents run through an SDK whose base URL you can redirect.
What to do
Setup and operation, verbatim from the page:
git clone https://github.com/muhamadgalihsaputra/shannon-proxy.git galyarder-security
cd galyarder-security
cp .env.example .envEdit .env (example values shown):
ANTHROPIC_BASE_URL=http://host.docker.internal:8317
ANTHROPIC_API_KEY=***
CLAUDE_MODEL=gemini-3-pro-previewBuild and run:
docker compose build --build-arg DEFAULT_MODEL=gemini-3-pro-preview
docker compose up -d
./shannon start URL=https://target.example.com REPO=/path/to/target/repoCleanup is destructive — preview first:
./shannon cleanup REPO=/path/to/target/repo DRY_RUN=true
./shannon cleanup REPO=/path/to/target/repoCaveats
- Provenance mismatch worth checking before you run it: despite the
galyarderlabs/galyarder-securitycatalog URL, the install instructions clone from an unrelated personal account (muhamadgalihsaputra/shannon-proxy). Verify what you are actually pulling. - The example config sets
CLAUDE_MODEL/DEFAULT_MODELtogemini-3-pro-previewwhile routing through Anthropic-named env vars — the SDK naming and the target model are decoupled, so confirm your proxy maps them as intended. - The page states the design goal ("zero-residue," "auditable evidence") but provides no validation of coverage; it explicitly frames itself against tool-using agents, not prompt-injection chat scenarios, so don't expect it to address the latter.
Technologies named here
- galyarder-security
- CLIProxyAPI
- LiteLLM
- OpenRouter
- Temporal
- Docker / Docker Compose
- Ollama
- shannon CLI
- Docker
- Anthropic
- Gemini
Techniques named here
- Auditable Proxy Routing
- Temporal Security Workflows
- Local Cleanup Routines
- Sovereign Gateways
- Auditable Evidence
- Vigilance over trust
Related in AI Security
