$ the-wire · showcase
Guardrails exposes hooks over HTTP, certifi floor raised in agent-scan
By RepoJournal · Filed · About Agent Security · Composed from the cited sources · methodology
NeMo Guardrails turns guardrails into an HTTP call your agent harness can make before any tool runs, and snyk/agent-scan raises its certifi floor for a Snyk-flagged certificate vulnerability.
feat(hooks): Add Hooks Server NVIDIA-NeMo/Guardrails
A FastAPI hooks server now answers PreToolUse calls at `/v1/hooks/claude-code`, `/v1/hooks/codex` and `/v1/hooks/nemo-relay`, so Claude Code, Codex and NeMo Relay harnesses can send each prospective tool call to your rail over HTTP/REST before it executes, without embedding the library. It sits on the Pydantic client models landed just before it.
feat(server): resolve trusted Rails configurations NVIDIA-NeMo/Guardrails
Declared Rails configurations are now preloaded before the application reports ready, and each buffered or streaming request selects an already-initialized checker based on trusted request context, provider instance and operation. That moves the rail-selection decision off provider composition and transport routing and onto a catalog of `ConfigEntry` IDs you declare up front.
fix: bump certifi to >=2026.10.19 for Snyk vulnerability snyk/agent-scan
The dependency requirement moves to certifi >=2026.10.19 to close a vulnerability flagged by Snyk. If you install agent-scan into a pinned environment, the old floor leaves you on the vulnerable bundle.
feat(server): add application-owned HTTP transport extensions NVIDIA-NeMo/Guardrails
Applications can now supply an `httpx.AsyncBaseTransport` factory that the assembly calls once per application lifespan and shares across provider instances, plus lifespan ownership of a shared HTTPX client. Model routing, gateways and custom network policy stay in your application rather than the guarded kernel.
fix(agent-runtime): report paused Docker containers as stopped, not failed (#4410) microsoft/agent-governance-toolkit
Elsewhere: `DockerDeployer.status` in the governance toolkit now maps `paused` and `removing` to STOPPED instead of FAILED, so an agent paused with `docker pause` no longer reads as a failed deployment, while `dead`, `created` and `restarting` still do.
Action items
- → Upgrade agent-scan's certifi pin to >=2026.10.19 before your next install snyk/agent-scan [immediate]