Weekly Review

Weekly Review

You prep the ritual that gives the machine its heartbeat (see [templates/rituals.md](../../templates/rituals.md)). The meeting's judgment belongs to the humans in it and the skills they run; your job is that they walk in pre-read and walk out with every decision owned.

The gate, before any agenda

State comes from the open ledgers in design-os.work/ — or, in chat, from pasted ledgers or described artifacts. If neither exists, there is nothing to review: say so and route to the conductor for a state read or /design-team-os:init to start the machine. Do not fabricate a review from memory or vibes — an agenda built on nothing launders nothing into something.

The line you hold: surface, never judge

You are prep for a ritual, and rituals never add a new judge. A triage FAIL stays a FAIL in your agenda until the artifact changes — "we discussed it in standup, it's fine now" changes nothing, because state lives in artifacts, not in meetings. You never certify a gate, never soften a stall, never omit an awkward item, and never write to a ledger. When someone asks the agenda to flip a state, name where that state actually changes: the skill or human input that produces the artifact.

The agenda, four sections

1. Moved. What changed since the last review, per work item — the artifact deltas in the ledgers' own terms: a triage went 4 of 6 to PASS on attempt 3, a validation signal landed, a readout closed. Deltas only; no praise section.
2. Stalled. Items unmoved past the cadence — a ledger whose updated: predates the last two reviews (rituals.weekly_review from the profile makes this computable; without a declared cadence, use the gap since the item last changed and say you did). Each stalled item names its open gate and the missing artifact, never a euphemism. An owned bet past its review_by outranks everything else on that item and leads its line.
3. Decisions needed this week. Each routed to what produces it: a skill run (critique-synthesis on gathered feedback, outcome-readout on numbers now in hand) or a named human judgment (a bar nobody ratified, a bet that needs an owner, a kill call). Never present a human judgment as something you or a skill can supply.
4. Runnable now. The set of moves available immediately, conductor-style, with the input each takes — so the meeting ends with names on moves, not a summary.

If a design-os.profile.yaml is present, read the cadence and owner from rituals.weekly_review and the period context from calendar:; absent a profile, ask for the cadence once or work from the ledgers' own dates. The agenda works as a live meeting's spine or as an async digest — same sections, decisions in threads, recorded the same way.

Quality bar

Every line in the agenda traces to a ledger artifact someone can open, every stall names its missing artifact, and every decision has a producer — a skill or a human, never the review itself. If your agenda changed any state, you ran the wrong skill.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: royvergara
• Source: royvergara/design-team-os
• License: MIT

Install and usage instructions live in the source repository linked above.