synthesize-digest

Synthesize digest

Purpose

A passive model only speaks when spoken to. It can hold a perfect picture of how a module works and still let that picture rot, because nobody thought to ask the right question this week. The whole point of a business ontology is to be a living model of reality — and a living model needs a heartbeat, not just a query interface.

This skill is that heartbeat. On a cadence, the agent looks across the whole model and surfaces what a human would actually want to know without having to dig: what got committed since last time, what is waiting on a human decision, where the model has drifted from reality, which metrics moved, and which recorded human requests are still open. It compresses all of that into one short, important-first message and posts it to the team channel.

The reasoning behind making this proactive rather than on-demand: drift and staged proposals are exactly the things nobody remembers to check. A staged card that sits unpromoted for three weeks is a fact the team agreed mattered and then forgot. An overdue next-audit is a card quietly going stale. A passive agent never raises these on its own, so they accumulate silently until the model is no longer trusted. A scheduled digest converts "silently rotting" into "visibly on the agenda," which is the difference between a model people rely on and a model people stop reading.

It is also a reporting skill, not an acceptance skill. The digest names what is staged and recommends what deserves attention, but it never promotes anything itself. The agent proposes; an authorized human decides; the deterministic controller applies. Surfacing a pending proposal is help; promoting it on agent authority would be self-approval.

This is a self-initiated skill. The agent reaches for it on its schedule, or the moment someone asks for the state of the module — not because every change needs an announcement, but because the rhythm is what keeps the model alive.

When to use

Reach for this skill when: