An evidence record beside deny_tool (typed and replayable)?

[Rul1an]

What stood out reading cascadeflow is that you run inside the agent loop rather than at the HTTP boundary, and you already make per-tool-call decisions there, deny_tool and the rest, and call them auditable decisions. That last part is exactly where I have been spending my time, from a slightly different angle: recording each privileged tool decision as a typed, replayable artifact, so a reviewer can pick it up later and replay it. I keep the boundaries honest because it is easy to overclaim, an allow is only the decision to forward and never proof the action happened, and a deny is fail-closed caution rather than a verdict on intent.

There is a one-minute demo if it is easier to look than to read: https://github.com/Rul1an/assay-agent-gate-demo has two open PRs, a red one where an agent tries a deploy key it was not granted (denied before it forwards, and the denial shows up in the Security tab) and a green one where the same action is declared and scoped. It runs offline against a mock, no real provider.

I wondered whether a record like that, emitted alongside a cascadeflow deny_tool or allow, would be useful to your users who already care about auditable decisions. Happy to send a small runnable example, and I would value your read either way.