Executable docs & execution security

[nayname]

Hi - I’ve been working with Mintlify-based docs in the Cosmos ecosystem (e.g., Juno, Neutron), exploring how documentation can move beyond answering questions to enabling user-initiated actions via the Mintlify Agent.

In financial or other complex environments, this introduces a clear “leap of trust” for users: once actions can be proposed by an LLM inside docs, questions around execution, confirmation, and responsibility become unavoidable.

As documentation platforms become more interactive, this feels like a problem the platform will eventually need to account for.

I’ve been exploring an external execution-layer approach that sits between the model and backend actions and focuses on deterministic planning and guardrails (intent → explicit execution plan → verification → action), rather than direct tool invocation. The goal is to reduce nondeterministic model behavior and make execution predictable before anything runs.

Curious whether this problem resonates here. Happy to share implementation details or get a sanity check on the architecture if useful.