loading
Loading content
loading
Vulnerabilities
Black Hat Arsenal reference monitor for coding agents, inventoried as a policy and provider corpus.
overview
GolemHalt is the Sondera coding-agent hook layer: a deterministic reference monitor that sits at the point where Claude Code, Cursor, Copilot, Gemini and similar agents call out, and decides with Cedar policy and YARA-X rules whether the call proceeds. Deterministic matters here, because a policy gate that asks a model for permission inherits the model's failure modes.
The Trickest node covers the inventory side of that monitor. mode=info reports the upstream Cedar and YARA policy corpus and the provider surface it supports; mode=hook-help with a provider explains the hook wiring for that agent. Promoting the full sondera binary is a separate step, so this node reads the policy layer rather than enforcing it.
Inventory is still worth scheduling. The corpus is what your enforcement depends on, and the Coding Agent Policy Gate runs golemhalt beside agentsleak so a change in the policy surface correlates with what the recorded event stream shows the agents actually doing.
use cases
Run info mode to record which policies and rules the upstream monitor ships, as data rather than as a document someone maintains by hand.
Read the provider surface to see which coding agents the hook layer supports before you assume a fleet is uniformly gated.
Use hook-help with a provider name for the wiring detail specific to that agent rather than the generic set.
Pair it with agentsleak so the policy corpus and the evaluated event stream land in the same scheduled report.
reference
| Name | Type | Flag | Description |
|---|---|---|---|
| mode | STRING | --mode | info (default) or hook-help |
| provider | STRING | --provider | Provider for mode=hook-help (claude, cursor, ...) |
Showing key inputs. golemhalt exposes 2 inputs in total.
example
# inventory the policy corpus and provider surfacegolemhalt --mode info # hook wiring for one agentgolemhalt --mode hook-help --provider claude{ "tool": "golemhalt", "mode": "info", "engine": "cedar", "policy": "deny-egress-after-credential-read", "providers": ["claude", "cursor", "copilot", "gemini"], "severity": "info", "note": "Corpus inventory only; enforcement happens at the agent hook"}guidance
Reach for golemhalt when you want the deterministic policy layer for coding agents inventoried and version tracked. Use agentsleak when you have recorded events to judge, and use semgrep-scan when the target is the code the agent produced rather than the agent's own calls.
Evaluates recorded agent-action events after the fact. golemhalt describes the monitor that would have decided at the hook.
Rules over source code. Use it on what the agent wrote; it has no view of what the agent was allowed to call.
Secret detection in repositories. Adjacent evidence for the same fleet, unrelated to policy enforcement.
workflows
faq
related
Black Hat Arsenal runtime security for AI coding agents, evaluated offline over recorded action events.
Automates OS command injection detection and exploitation.
Parameter mining and XSS testing with headless verification.
Subdomain takeover scanner with cloud-zone intake.
Locate public CVE proof-of-concept repositories on GitHub.
Black Hat Arsenal deception layer for MCP servers, exported as a schedulable ghost-tool catalog.
A mode and optional provider feed golemhalt, which writes the Cedar and YARA policy corpus as findings for correlation.
Facts on this page come from the live Trickest tool library.