Idea: Native Causal Auditing (K9) integration for safer agent executions

[liuhaotian2024-prog]

Hey team, love the velocity of the project!

I notice many users deploying agents eventually hit the "silent deviation" wall—where the agent technically succeeds (exit 0) but conceptually fails (e.g., writes to the wrong env).

I maintain K9 Audit, a causal conformance auditing ledger (AGPL-3.0). It tracks agent actions via a hashed CIEU 5-tuple without using LLM-as-a-judge. I’ve already mapped out a module-level wrapper (k9_wrap_module) that works seamlessly with OpenClaw skills.

Would you be open to a PR adding a lightweight, optional K9 Audit hook into your examples or core framework? It gives your users instant, tamper-proof audit trails for enterprise compliance.
Let me know your thoughts!

[vitas]

hi, do you want contribute to my project? https://github.com/vitas/evidra
it could be integration also for gstack

[liuhaotian2024-prog]

Hi — thanks for proposing an integration between gstack and K9 Audit.

Before we start a PR, I want to confirm a few design points so we align:

1. Should the integration be an example usage in gstack docs, or a reusable hook/plugin?
2. When should K9 Audit run — after each agent step, on completion, or only on error?
3. Do you expect default constraint settings, or user‑configurable rules?

Once clarified, I can prepare a minimal K9 Audit prototype for gstack:

• Add an audit step in gstack workflows,
• Demonstrate trace & causal report output,
• Show constraints enforcement.

Appreciate the interest!