This guide explains the design principles for Moiravo’s planned execution workflow. The current local thread and context foundation does not provide a complete agent-execution or enforcement system.
A conversation is not permission.
An idea can be worth exploring before anyone is ready to act on it. Linking a conversation, importing its history, or asking a question should not silently grant an agent access to systems or permission to make changes.
We are designing execution around explicit scope: the outcome being pursued, the resources that may be used, the actions that are allowed, and the limits that apply. Expanding those bounds should require a new decision from the person who owns them.
Keep consequential choices legible.
Some steps gather information. Others publish, spend, deploy, delete, or change access. A useful workflow makes the difference visible at the point of action.
The intended experience ties consequential actions to a specific approval and a clear result. Any integration’s actual controls must be demonstrated before the product can claim that those bounds are enforced. A setting or a stop request alone is not proof that an external process has stopped.
A result should point to its basis.
“Done” is helpful only when it means something concrete. For a code change, the evidence might be the exact changed files and the checks that ran. For research, it might be the sources reviewed, their dates, and the questions they could not answer.
We distinguish the kind of evidence from the way it was reviewed. A human reviewing a simulation does not turn it into an observed event. An agent’s confident description does not establish that a deployment succeeded.
Uncertainty belongs in the record.
When a tool disconnects during an external action, the outcome may be unknown. The right next step is to determine what happened before repeating the action. A missing receipt should stay visibly unresolved.
The same principle applies to imported history and source freshness. Preserve the useful material, state its limits, and avoid upgrading an assumption into a fact simply because it appears in a dashboard.
The ambition is less supervision, with better grounds.
Clear scope and evidence should help an operator spend attention on meaningful decisions. They should also make it possible to compare tools on useful results rather than activity alone.
These are product requirements we still have to prove in the connected workflow. Read the security approach and development status for the current boundary.