Agent Governance Evidence Pack | Business-facing governance evidence for AI agents
Evidence/Agent Governance Evidence Pack

Agent Governance Evidence Pack

Business-facing evidence packages for governed AI-agent deployment.

A lightweight public schema, validator, and markdown renderer for summarizing AI-agent runtime governance evidence — what an agent is deployed to do, which tools and systems it can access, which actions require authority or review, what was blocked, what sources were relied on, what replay bundles exist, what validation results are available, what exports were redacted or signed, and what risks remain open.

Apache 2.0 Evidence-pack schema Summary generator Markdown renderer Templates Tests
Agent evidence pack
Deployment readiness summary
Overview
Tools & actions
Policy controls
Replay bundles
Risk register2 open
Review records
The problem

Governance teams do not review raw traces.

Security, risk, compliance, audit, and executive teams need business-readable evidence. Raw logs and technical traces are not enough. They need a structured summary of the agent's purpose, deployment context, action surface, authority model, controls, blocked actions, reliance records, replay bundles, validation results, redacted exports, risks, and review history.

01

Technical records are hard to review

Raw run records and traces assume engineering context that governance and executive reviewers do not have.

02

Evidence is scattered across systems

Manifests, control-plane records, and replay bundles live in different systems with no single review surface.

03

Risk owners need business-readable summaries

Risk and compliance owners need plain-language summaries tied to controls, not raw records they must interpret.

04

Approval decisions lack consistent artifacts

Deployment approvals are recorded inconsistently, so there is no standard artifact behind a go or no-go.

05

Audit review needs structured references

Audits need stable references from findings back to the underlying runtime evidence.

06

Deployment readiness is hard to summarize

There is no consistent way to summarize whether an agent is ready to deploy and what risks remain open.

What it does

Turns runtime records into governance evidence.

A

Agent overview

A plain-language description of what the agent is deployed to do.

B

Deployment context

Where and how the agent runs, and which systems it touches.

C

Tool inventory

The tools the agent can access, summarized for review.

D

Action inventory

The actions the agent may propose, with their types.

E

Authority model summary

How authorization is required and granted across actions.

F

Policy control summary

The runtime controls in place and what they allow or block.

G

Blocked-action summary

What was blocked during operation, and why.

H

Reliance summary

The sources the agent relied on across recorded runs.

I

Replay bundle inventory

Which replay bundles exist and what they cover.

J

Validation summary

The validation results available for the recorded evidence.

K

Redaction export summary

What was redacted in the exported evidence, and how.

L

Risk register

Open and resolved risks associated with the deployment.

M

Review records

Who reviewed the agent, when, and with what outcome.

N

Markdown rendering

Renders the pack to markdown for review and distribution.

Why it is valuable

A bridge between platform teams and governance stakeholders.

Helps executives understand agent deployment posture
Gives risk and compliance teams a consistent review artifact
Connects technical runtime records to business controls
Summarizes blocked actions, reliance, validation, redaction, and signing
Supports incident review and deployment readiness discussions
Bridges AI platform teams and governance stakeholders
How it fits in the stack

The Evidence layer of the Open Control Stack.

Agent Governance Evidence Pack is the Evidence layer. It translates manifests, control-plane run records, and replay bundles into business-facing review material for governance, risk, security, compliance, audit, and executive stakeholders.

Declare Control Replay Evidence
What it is not

Evidence — not a conclusion.

×Not a compliance certification
×Not a legal conclusion
×Not an agent framework
×Not a model runtime
×Not a hosted GRC platform
×Not proof that model outputs are correct
×Not a replacement for legal, risk, or compliance review
Open-source contents

What is in the repository.

Evidence-pack schema
Pydantic object model
Validator
Summary generator
Markdown renderer
Templates
Example evidence packs
Tests
Documentation
View on GitHub
Who it is for

Built for the people who approve agent deployments.

CISOs and risk leaders

Need a business-readable summary of an agent's deployment posture and open risks.

Governance and compliance teams

Need a consistent review artifact tying runtime evidence to business controls.

Executives and approvers

Need to understand what an agent does and whether it is ready to deploy.

Auditors and reviewers

Need structured references from the summary back to underlying runtime evidence.

AI platform teams

Need to translate runtime records into material governance stakeholders can review.

Get started

Turn agent runtime records into reviewable governance evidence.

View on GitHub