Repository navigation
Feedback on Agentmetry interoperability and useful AGT validation work #4560
Replies: 2 comments
|
Not a maintainer, but I read your adapter ( 1. A policy decision is imported as a tool call. 2. Without the key, the chain check is not tamper evidence. The HMAC is symmetric, and Also worth knowing: On question 2: start with fixtures, because the signed file is the artifact a third party can check, and none of the 12
— John B Joseph, Co-founder & CPTO, Purplecop Security |
|
Thanks, John. I checked both paths and confirmed the mapping and verification issues. I’ll start with a fixture generated by govern() that imports as two policy decisions and zero tool calls. I’ll also make the importer report chain consistency and HMAC verification separately. Your approval and rotation cases give us a concrete follow-up for the fixture set. I’ll share the results here once I’ve run them. |
Uh oh!
There was an error while loading. Please reload this page.
I maintain Agentmetry, a local sensor for AI coding agents. Following the earlier audit-format contribution and evaluation/pilot documentation, I would appreciate feedback from AGT maintainers or users on the project and its integration boundary.
Agentmetry has a
FileAuditSinkfile importer. I have also been revisiting the feedback on #3732, including isolating fixture generation from ambient environment variables and writing regenerated fixtures undertmp_path.Before refreshing that work, two questions would help me choose a useful scope:
GovernanceEventSinkinterface, be more useful to AGT users? A concrete scenario you would want tested would be especially helpful.Any example would need to preserve source IDs, policy verdicts and provenance, while keeping a policy decision distinct from evidence of tool execution. I am asking about the useful contribution boundary here, not claiming current live compatibility for a new adapter.
A few pointers would help me prioritize as a solo maintainer. Thank you.
All reactions