Decisions die in the gap between the meeting and the tracker. The notes capture 'Sarah will own the migration plan' and then nothing happens, because transcribing commitments into tickets is low-status work that loses to everything else. Two weeks later the same meeting re-litigates the same decision.
The gap is small — minutes of clerical work — but it's recurring, boring, and structurally unowned, which is why it never closes.
A Fleet workflow reads committed meeting notes, extracts the commitments, and — after the meeting owner approves the extraction — files them as real tickets via the workflow's ticket output.
genflows:
- name: action-items
steps:
- {name: extract, prompt: "Extract action items from the notes: owner, commitment, due date if stated. Verbatim quotes for each. Flag ambiguous ownership.", corpus: ["meetings/notes/*.md"], kind: report, out: actions.md}
- {name: owner-ok, depends_on: [extract], kind: approval, out: decision.md}
In the dashboard builder, the post-approval step's output sink is a ticket: each approved action item becomes an issue with the owner and the quoted commitment, created through Fleet's ticket-system integration (GitHub issues today). The approval gate matters here precisely because tickets ping people — the meeting owner confirms the extraction before anyone gets assigned anything.
Notes corpus → extraction with verbatim quotes → owner approval → tickets created. The clerical gap between 'we decided' and 'it's tracked' closes structurally.
Chiefs of staff, EMs, and ops leads who run decision-heavy meetings and keep notes in a repo (or can export them to one) — anywhere follow-through depends on a human remembering to transcribe.
The corpus is file-based, so the notes need to land in a repo — many teams already export or sync them. The extraction quality tracks the notes quality: structured notes with names produce clean items.
GitHub issues today, through Fleet's ticket-system abstraction. The workflow creates or refines issues as its output sink — assignment and labels included.
One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.