Work through decisions with AI
Turn feedback, meeting notes or research into one decision at a time, keeping accepted changes, deferrals and missing evidence visible in the discussion.
Work through decisions with AI when a review ends with pages of comments and several competing ideas about what the team agreed. The decision-walkthrough skill asks an agent to read the authorised material first. It identifies which choices depend on earlier ones and brings you one item at a time, with the source, a recommendation and the main trade-off.
If the audience for a page is unsettled, choosing the final button wording may be premature. The earlier decision can come first. If a participant challenges the agent’s interpretation, the original wording can be checked before a decision is recorded. An uncertain response can remain open.
What it does
- Separate a direct quotation from an inference or material assumption.
- Group repeated feedback without losing a distinct objection.
- Order choices by dependency before the team addresses individual details.
- Mark an item as needing evidence and identify a reasonable way to obtain it.
- Reconcile agreed, revised and deferred items against the original source.
Install and use
- Provide the authorised notes or research and specify whether you want exhaustive coverage or only choices that affect action. No external service is required.
- Ask, “Walk me through this review one decision at a time. For each one, show the source and wait for my choice.”
- Review the final log and ordered implementation checklist before authorising any work.
FAQ
Source
Made by Jarad Johnson. Original skill: decision-walkthrough, published on GitHub under the MIT licence.
View on GitHub