policy-testing

LLM guardrails & prompt injection detection for Python. Auto-instruments LangChain, CrewAI, OpenAI, LiteLLM + 8 more frameworks. PII masking, toxicity detection, policy CI/CD. One line, zero code changes.

Headless, agent-driven test harness for NetIQ/OpenText Identity Manager (DirXML) channel policies — runs the real engine, steps the chain, mines traces for inputs.

AuthContract turns institutional rules into developer-usable, testable controls. It connects authoritative rule sources to canonical contracts, CI/CD merge gates, runtime action checks, and reconstructable evidence—so teams can prove not just what an agent did, but which rule authorized it, what version applied, and why execution was allowed.

Deterministic CI regression testing for policy drift across model-visible tools, semantic validation, SQL compilation, database containment, and result release.

Red-team and compliance test harness for LLM outputs -- detects PII leakage, secret exposure, and policy violations with CI-friendly reporting.

Measure a coding-agent command hook: replay a corpus through two versions and diff the decisions. Decision replay, not a sandbox.

SignWarden: early research on receipts and audit records for AI agent actions.

Replay identical requests against OpenBao policy revisions and expose changed access in HTML, JSON and CI reports.

Behavioral testing for AI agents: define allowed behavior, run unchanged agents, and get policy-aware audit reports

A Go CLI tool for testing AWS IAM policies before deployment using scenario-based YAML configurations and the SimulateCustomPolicy API

Dry-run AI agent policy changes against historical traces before they go live — verdict deltas, regression reports, CI exit codes. Python, stdlib only.

Pre-deployment policy test harness for automated decisions, with adversarial cases and value-at-risk reporting.

Test your application's YorvoID authorization policy: write the decisions it must give as PHP cases and run them in PHPUnit or from the command line, against the policy as production builds it, with in-memory stores.

Check saved Hindsight bank route responses against allow and deny policies

Tests the plain-English rules that govern AI agents with a wallet. Paste a rule and Lacuna attacks it eleven ways, then reports what it actually permits, quoting your own words back. Audits a live agent over HTTP against its real Kite spending caps. Built on OpenServ SERV.

GhostGuard Policy Replay + Guard Score — backtest a new policy against a recorded trail and score trail quality. By Ghost Developer Studio.

Replay changed DecisionPacks over historical data, reuse judgments and measure decision flips.