Detector harness
Third-party detectors wired through the primust-hook subprocess framework. Each fires on PreToolUse / PostToolUse in Claude Code and Cursor, and its verdict is logged as a check execution record with provider attribution. Detector policy is managed here, served by the API, and refreshed by the hook on the next session start with a 5-minute cache. The local JSON file remains a fallback if the API is unavailable.
Loading detectors…
Available to wire
Detectors in the known universe that fit the harness pattern but aren't wired yet. Pick one and open an issue to request it — or follow the provider link to evaluate it yourself first.
Rebuffcloud_or_local
Multi-layer prompt injection (heuristic + vector + LLM)
Rebuff.ai·Apache-2.0
Llama Guard 3local
Input/output safety across 14 hazard categories (7B)
Meta·Llama 3 Community License
NeMo Guardrailslocal
Jailbreak + dialog policy + fact grounding
NVIDIA·Apache-2.0
Guardrails Hublocal
60+ validators: PII, toxicity, JSON, jailbreak, …
Guardrails AI·Apache-2.0
Perspective APIcloud
Toxicity, insult, threat, identity attack scores
Google Jigsaw·Free (rate-limited)
Socketcloud
Malicious npm/PyPI package detection via behavioral analysis
Socket.dev·Free tier for public packages
Snyk Codecloud
SAST with ML for JS/TS/Python/Java
Snyk·Free tier for open source
CodeQLlocal
Semantic code analysis (SAST) via query language
GitHub·Free for open source
Trivylocal
Container, IaC, dependency & secret scanning
Aqua Security·Apache-2.0
Grypelocal
Vulnerability scanner for containers & packages
Anchore·Apache-2.0
HiddenLayer Model Scannercloud
Malicious ML model artifact detection
HiddenLayer·Commercial
BigIDcloud
Enterprise data discovery + PII classification
BigID·Commercial
Private AIcloud_or_local
54-language PII detection & redaction (on-prem available)
Private AI·Commercial
DeepKeepcloud
LLM jailbreak + data-leak runtime guard
DeepKeep·Commercial
CalypsoAIcloud
Enterprise GenAI gateway with prompt/response scanning
CalypsoAI·Commercial