Skip to main content

04.22Prioritisation

Strategic alignment: Claude Code skill for prioritisation

Claude Code skill for prioritisation, built at AR Labs: when focus breaks down, list active projects, sort them by value and momentum, and pick one thing.

Engraved plate: a mariner’s compass with project markers around the bezel, its needle on one starred mark: the next thing.An engraved mariner’s compass in a gimbal ring; scattered markers for active projects sit around the bezel, and the needle points to a single starred mark, the one thing to do next.AR LABS · INSTRUMENT CABINET · DRAWER 04CALL NO. 04.220123456789NABCDFIG. 22 · STRATEGIC ALIGNMENTMARINER’SCOMPASSPARTSACheck thedirection holdsBEvery activeproject, inventoriedCEach sorted byvalue and momentumDThe needle:one thing nextRECOVERS FOCUS
On this page
  1. How it works
  2. The judgement it encodes
  3. What exists today
  4. Example

Strategic alignment is a Claude Code skill, built at AR Labs, for recovering focus. It is for anyone running several projects with AI agents. When too many projects are open at once, a schedule has fallen apart, or it is no longer clear what matters most, an agent runs this six-phase session with the person whose work it is. The agent confirms the overall direction still holds, takes stock of every piece of active work, sorts each project by its value and momentum, and narrows the field to one thing to work on next. It ends by loading everything needed to start that one thing.

How it works

FIG. 01
flowchart TD
  P1["1. Congruence check"] --> P2["2. State assessment"]
  P2 --> P3["3. Pivot analysis"]
  P3 --> P4["4. Priority resolution"]
  P4 --> P5["5. Context loading"]
  P5 --> P6["6. Handoff to execution"]

1. Congruence check. Read the vision, mission, current goals and the strategy agent's state, and ask whether they still hold. Confirming the existing direction is the agent's call; changing it is the person's.

2. State assessment. Query the task tracker (we use Beads, Steve Yegge's open-source issue tracker for coding agents, which lives in the repository) for everything in progress or blocked, and build one table: project, goal, status, last activity, momentum and value. Count the items in progress as an attention-fragmentation score: one or two is focused, three or four is borderline, five or more is critically fragmented.

3. Pivot analysis. Place each project in a two-by-two matrix of value against momentum. The agent may mark a project to accelerate on its own; unblocking, redirecting or stopping a project needs the person's approval.

High momentum Low momentum
High value Accelerate Support: unblock it
Low value Pivot: redirect it Stop

4. Priority resolution. Take the projects to accelerate and the approved ones to support. Weight them by the current goal they serve, and score each from 1 to 10 on several dimensions, among them learning value, how much other work it unblocks, how close it is to done, and its value as visible evidence of the work. Present the top three with reasons. The person chooses the one thing, or overrides.

5. Context loading. Read the chosen project's guide, design documents and tracker notes, and condense them into a context package: what, why, where it stands, how, and the concrete next action. Then write "bridge cues" that carry the person from a scattered state into focused work: where they were, what the next step is, and why it matters.

6. Handoff. Update the current-focus record if the focus has shifted, raise the chosen task to top priority in the tracker and mark it in progress, save the session's tables and reasoning as a dated record, break the next step into three to five concrete actions, and move from strategy into execution.

The judgement it encodes

Count the open loops. Fragmentation is measured, not felt. A number of active items is harder to argue with than a sense of being busy.

Momentum and value are separate questions. A high-value project with no momentum needs its blocker found and removed. A low-value project with momentum is redirected, not simply allowed to continue because it is moving.

The agent calculates; the person chooses. Scoring and presenting the top three are delegated. Selecting the one thing, and any change to direction, are not.

End ready to work. A prioritisation session that ends with a decision but no loaded context leaves getting started undone.

Know when not to use it. A question about today's work is answered from the tracker's ready list, a calendar conflict goes to the scheduler, and being stuck on one task means debugging the blocker, not re-examining the whole portfolio.

What exists today

The skill is part of our agent setup and is run on request, when focus breaks down. It sits beside the strategy agent's regular planning cadences, which keep direction current on a schedule, and the scheduler, which handles time rather than direction.

Example

Illustrative example. The state summary from phase 2:

| Project            | Status      | Last activity | Momentum | Value |
|--------------------|-------------|---------------|----------|-------|
| Voice interview UI | in progress | today         | High     | High  |
| Skills pages       | in progress | 2 days ago    | Medium   | High  |
| Scheduler refactor | in progress | 3 weeks ago   | Low      | Low   |
| Diagram theme port | blocked     | 1 week ago    | Low      | High  |
| Tooling experiment | in progress | yesterday     | High     | Low   |
 
Attention fragmentation: 4 in progress (borderline).
Low momentum, high value: Diagram theme port (support: unblock).
Low momentum, low value: Scheduler refactor (candidate to stop).