---
name: "skill-goal-review"
type: skill
description: Periodic review of progress against the outcome-initiatives and their KPIs across all ventures. Use this whenever running the goal-review routine, or when Aled asks how the strategy is tracking, which goals have stalled, or which KPIs are off target. Reads initiatives, their linked tickets and milestones, and KPI movement, then reports what is on track, what has stalled, and what is drifting — and proposes refocus, de-scope, or escalation for Aled to decide. Read-mostly; proposes only. Produces a scannable review summary.
license: CC BY-NC-SA 4.0
---

# goal-review

Atlas's periodic strategy check. Roughly monthly or quarterly, it reads the outcome-initiatives across every venture and reports honestly on where the strategy stands — on track, stalled, or drifting — and proposes what to do about it. It does not change the strategy itself; that is skill-strategy-cascade with Aled's approval. Writes to Linear only; reads the KPI sources the cascade names, read-only, because Linear holds the targets and not the numbers. Follows `convention-linear` and `core-operating-model` throughout.

## Trigger

Scheduled run roughly monthly or quarterly across all ventures: careerOS, Apps, A1, Mr Pritchard, and madebyneud. Also on demand when Aled asks how the goals are tracking. Never touches the Pipeline team.

## Behaviour

A run has three passes, in order.

### Pass 1 — Read the goals layer

Read all outcome-initiatives across the ventures (the source produced by `skill-strategy-cascade`). For each, note: name, strategy statement, objectives, KPI targets and baselines, current status, linked tickets, and milestone targets.

If no outcome-initiatives exist for a venture, note the gap in the summary and propose running `skill-strategy-cascade` for that domain rather than reviewing against nothing.

### Pass 2 — Assess movement

For each initiative, read the linked tickets and milestones and the KPI movement since the last review, and place it in one of three states:

- **On track** — KPIs moving toward target at a reasonable pace; milestones being met.
- **Stalled** — no meaningful progress in the period: no tickets closed, milestones slipping with no activity, the initiative quietly parked.
- **Drifting** — work is happening but the KPIs are off target, or the activity no longer maps cleanly to the objective it was meant to serve.

Be honest about stalls and drift. A clear "this has not moved" is more useful to Aled than a generous reading.

### Pass 3 — Report and propose

Produce the review summary (below). For each initiative that has stalled or is drifting, propose one of:

- **Refocus** — the goal stands but the effort is misdirected; recommend where attention should move.
- **De-scope** — the goal is too large or no longer worth its full ambition; recommend trimming it.
- **Escalate** — the goal is blocked or at risk in a way Aled needs to decide on directly.

Each proposal is a **line in** `osclaude.goal-review`, in the review-summary format below, whose `[link]` resolves to the existing ticket that carries the decision. That is what "posted for Aled" means here: posted where Aled reads it, which this skill names as that document. **A proposal becomes a new ticket only where no existing ticket carries it and the proposal names an action rather than a judgement** — and then it is filed once, with the goal-review line pointing at it, never re-filed when the same proposal recurs on a later run. The recurrence guard is not optional at a weekly cadence: without it the same set is re-filed indefinitely, which is the comment-idempotency defect of APP-384 and APP-738 in ticket form. Do not change initiatives, cancel goals, or re-cut strategy in this skill — those are proposals Aled approves, then actioned via `skill-strategy-cascade`. The reading matters because this skill is read-mostly and proposes only: on 2026-09-14 the run produced six proposals while its own headline finding was that roughly 69 items already waited on the Owner, the oldest at 61 days, so the alternative reading would have answered "the Owner queue is the constraint" by adding six tickets to it (APP-1419).

## Review summary

Post the summary as one dated entry in the standing document **`osclaude.goal-review`** on the os.Claude project — one entry per run, newest first, keeping the `[goal-review]` marker line exactly as below. One document, one writer. It is the deliverable Aled reads:

**This skill creates the document where it does not exist**, as its first write of the run, before posting the entry — the same ordering `task-cycle-planning` uses for the cycle-goal document and for the same reason (APP-820, APP-987). **Entries written before the destination was named are not migrated**: a read-mostly skill does not move the Owner's prior records on its own read of a destination change, so earlier entries stay where they were posted and a migration, if wanted, is filed separately. A consumer testing freshness against this document therefore distinguishes *absent because never created* from *stale*, and reports the former as "the review has not run", never as nothing to do — inferring absence from a read known to be partial is the APP-895 / APP-1045 shape and is precisely what `task-strategy-review`'s Pass 2 fallback exists to absorb (APP-501, APP-1419).

The destination is named rather than offered, because "the os.Claude project **or** a designated strategy-review issue" designated nothing: three consecutive runs posted as a project comment, a convention no artefact recorded, so each run had to rediscover it — and the 2026-09-08 run's rediscovery cost a `list_comments(projectId)` read that overflowed at 51KB and needed out-of-band parsing (APP-1180). A named document also gives `task-strategy-review`'s Pass 2 and any freshness test a single artefact to resolve, instead of a comment thread to page through (APP-1179). This is the fix APP-1054 applied to `task-cycle-planning`'s commit proposal, in the same shape.

```
[goal-review] Strategy — [date] — [period covered]

On track:
• [venture] / [initiative] — [KPI] at [value] vs [target]

Stalled (no progress this period):
• [venture] / [initiative] — [what has not moved]

Drifting (off target):
• [venture] / [initiative] — [KPI] at [value] vs [target], [why]

Proposals (needs your decision):
? Refocus / De-scope / Escalate — [initiative] — [recommendation and reason] — [link]
```

Keep it scannable. Aled reads it at his review, not line by line.

## Guardrails

- Read-mostly. Proposes only; never changes strategy, cancels a goal, or re-cuts a cascade — those flow through `skill-strategy-cascade` on Aled's approval.
- Never marks an initiative or ticket Done.
- **Linear for everything it writes; read-only elsewhere for KPI actuals.** Linear holds the targets; it does not hold the numbers. Pass 2 cannot be run from Linear alone, so this skill **reads** the KPI sources named in the cascade — Shopify Analytics today, and whichever analytics, revenue or list source a future objective names — strictly read-only, and states every non-Linear source it read in the summary's sources line. The **write side does not widen at all**: it writes only to Linear, never code, never a PR, never the delivery loop, and never a write of any kind to a KPI source. Declaring the scope narrower than the job is what forced two consecutive runs to decide their own scope, the second one's only warrant being the first one's precedent (2026-08-22 and 2026-09-08; the 2026-09-08 run's eight-week Shopify series surfaced the review's single most material finding, so the excursion earned its keep — APP-1180).
- Never touches the Pipeline team.
- Honest assessment over generous: name stalls and drift plainly.
- Aled is the gate. Atlas proposes; Aled decides.

## Setup

- Deployed as a **Cowork scheduled task** (cloud) — the loader is the canonical line in `core-operating-model` (*The loader line — one form*), and `task-goal-review` is the artefact it reads. **Connectors: Linear (read and write), plus read-only access to the KPI sources the cascade names (Shopify Analytics today).**
- Cadence: roughly monthly or quarterly (or on demand when Aled asks how the goals are tracking).

Owned by Atlas (ATL-4S).

On finish, run skill-ops-retro capture on this run's friction: **search the Pulse queue first and add the evidence to a same-root note if one exists, otherwise file a new FRICTION note** for the drain pass to assess, in capture step 3's literal shape — team Apps, project os.Claude, `FRICTION:` title, labels `task` + `work:configuration` + `PUL-5E`, Backlog, unassigned — filing the friction, never the run report.
