ucsandman/dashclaw-guardrails

If you're running AI agents that can take real actions like sending emails or messages, you probably worry about them doing something they shouldn't. dashclaw-guardrails helps solve that problem by letting you write safety rules in a simple text file and then automatically turning those rules into actual tests that verify your agents behave correctly. Think of it as writing a checklist of "things my AI should never do without permission" and getting both automated tests and compliance reports from that single checklist. The workflow is straightforward. You write a policy file that describes what your agents are allowed to do, for example "any message send requires human approval." The tool then reads that file and generates a set of test cases that check whether those rules are actually being followed. You can run those tests locally or have them run automatically every time someone opens a pull request on GitHub. Alongside the tests, it produces a proof report in plain text and structured data formats that you can hand to a compliance team, auditor, or stakeholder who wants evidence that your guardrails are in place and working. This would appeal to a few different people. A startup deploying its first AI agent might use the "startup-growth" policy pack to get reasonable safety rules without starting from scratch. A larger company in a regulated industry could use the "enterprise-strict" pack and lean on the generated proof reports for audit purposes. A product manager might appreciate that the rules are written in readable YAML rather than buried in code, so you can see at a glance what your agents are and aren't allowed to do. One notable design choice is that the tool can either read policies from a local file or connect directly to a running DashClaw instance to pull live policies via an API. The README doesn't explain exactly what DashClaw is as a product, but the integration suggests it's a platform for managing agent policies centrally. The project currently generates tests for JavaScript projects using Jest, with Python test generation listed as upcoming. It also notes that more advanced rules like rate limits and data boundaries are still on the roadmap.