docs(integrations): add Agent Threat Rules (ATR) guardrail plugin#1850
Conversation
Adds docs/integrations/atr-guardrail.md (+ icon) for the adk-atr-guardrail package -- an in-process, deterministic security guardrail that enforces the open-source Agent Threat Rules ruleset at three fail-closed Runner callbacks (before_run / before_model / before_tool). Tagged 'resilience'. Created per maintainer guidance on google/adk-python#6130 to ship as a standalone package listed in the integrations catalog. Verified locally with mkdocs build.
✅ Deploy Preview for adk-docs-preview ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
|
Context for reviewers: this page was created at the request of @rohityan, who closed the original adk-python sample PR (google/adk-python#6130) with the guidance to "release the feature as a standalone package that we will then share through" the integrations catalog. The package is now published on PyPI ( It follows the same structure as the existing Cisco AI Defense integration page (#1672) — a guardrail plugin tagged |
… 2.0, and remove the resilience catalog tag
koverholt
left a comment
koverholt
left a comment
There was a problem hiding this comment.
Thanks for the PR and for contributing this integration! I made some small edits to switch to the App plugin pattern (avoids the deprecated plugins= warning), bumped the prereq to ADK >= 2.0.0, added model="gemini-flash-latest", and dropped the resilience tag. LGTM, merging.
3 participants
This adds an integrations catalog page for
adk-atr-guardrail, a standalone, in-process security guardrail plugin for ADK backed by the open-source Agent Threat Rules (ATR) ruleset.This is the follow-up requested in google/adk-python#6130: that PR added the guardrail as an ADK sample, and the maintainer recommended shipping it instead as a standalone package listed in the integrations catalog. The package is now published, so the page documents installing and using it.
What it provides
BasePluginthat enforces ATR detection at three fail-closed Runner callbacks: before_run (halts the run on a malicious user message), before_model (skips the model call when the assembled prompt still carries a threat), and before_tool (returns an error instead of executing a tool whose arguments match a rule).Details
resilience, mirroring the structure of the existing cisco-ai-defense page.Disclosure: I maintain the open-source ATR ruleset referenced here.