Repository navigation
Releases
AgentGov Auditor v0.1.0-alpha.1: governance checks for AI agents
First alpha release of AgentGov Auditor, a free command-line tool that scans a folder for AI agents and tells you which of 24 governance controls have real evidence behind them. It then gives one release decision: ALLOW, REVIEW REQUIRED, or BLOCK.
Scans are read-only and offline. No account, no API key, no telemetry, and the tool never runs the code it scans.
Highlights
- 24 governance controls in six domains, each with a stable ID. Examples: "a person approves risky actions" and "there is a way to stop the agent".
- Evidence levels, not a single green tick. Each control is rated from "declared only" to "verified in operation". A static scan alone reaches level 2 at most.
- Three risk profiles:
baseline,standard, andhigh-risk. A high score never overrides a failed mandatory control. - Framework awareness for LangChain, LangGraph, OpenAI Agents SDK, CrewAI, AutoGen, PydanticAI, MCP servers and clients, and generic tool calling, in Python, TypeScript, and JavaScript.
- Five report formats: terminal, JSON, Markdown, HTML, and SARIF.
- Signed evidence:
evidence digest|keygen|sign|verifybinds external test evidence to the exact files it covers, signed with Ed25519. - GitHub Action that builds the CLI from source, scans the repository, and can upload SARIF and post a pull request summary.
- Self-hosted dashboard (beta) with
agentgov serve, bound to loopback by default, with SQLite storage.
Install
Download the binary for your system from the assets below and check it against the checksums file. Or install with Go:
go install github.com/shahzadarain/agent-governance-auditor/cmd/agentgov@v0.1.0-alpha.1
Then scan a project:
agentgov scan path/to/your/project
Use it in GitHub Actions
- uses: shahzadarain/agent-governance-auditor@v0.1.0-alpha.1
with:
path: .
fail_on: blockFor real use, pin a full commit SHA that you have reviewed.
Alpha status
This is an alpha. While the version is 0.x, breaking changes can happen in any release. See docs/IMPLEMENTATION_STATUS.md for what has been verified and docs/known-limitations.md for what the tool does not do. A good result is not a certification that an agent is safe.
Full details are in CHANGELOG.md.
Full Changelog: https://github.com/shahzadarain/agent-governance-auditor/commits/v0.1.0-alpha.1