---
name: skill-prioritise
type: skill
description: >-
  Turn findings, themes, and opportunities into a scored, sequenced plan — apply
  the prioritisation standard (severity, or Impact × Effort × Confidence with the
  confidence overlay), output a ranked table and an Impact × Effort matrix, and
  split research-first items into their own track. Use this whenever Sonar needs
  to decide what to do next and in what order, after an audit or diagnostic. Reads
  standard-prioritisation; consumes the output of skill-discovery-audit and
  skill-conversion-diagnostic. Surfaced by the a1.Muse engagement.
---

# prioritise

Sonar's sequencing skill. Takes the clustered output of an audit or diagnostic and turns it into a defensible order of work, with the guesses pulled out into research rather than buried in the plan.

## Trigger

An audit or diagnostic has produced themes and opportunities, and the question is "what do we do, in what order?" Run after `skill-discovery-audit` or `skill-conversion-diagnostic`, before any roadmap is drawn.

## Behaviour

1. **Gather the inputs.** Collect the Theme Summaries and Opportunities (with their severity and confidence tags) from the audit or diagnostic.
2. **Choose the method.** Per `standard-prioritisation`: severity tiers for triage and gating; Impact × Effort × Confidence for sequencing interventions. Don't mix methods on one item.
3. **Score by deriving, not asserting.** Apply the chosen rubric, deriving each axis from its definition so the score is defensible. Check the severity accumulation rule where Highs cluster.
4. **Split the research-first track.** Pull high-stakes items tagged *unknown* into a separate research track per the confidence overlay — they become questions to answer, not work to commit.
5. **Output two views.** A **ranked table** (item, score, severity/confidence, track) and an **Impact × Effort 2×2** that makes quick wins, big bets, fill-ins, and time-sinks visible at a glance.
6. **Frame for the decision.** Present the order as a recommendation with its reasoning and the trade-off, so Aled (or the client) can decide. The skill sequences; it does not commit.

## Reads

`standard-prioritisation` (both methods, the confidence overlay), `convention-audit` (the Initiative card the ranked items become).

## Guardrails

Derive every axis from the rubric — a score you can't explain fails review. Keep severity and IEC apart per item. Always split high-stakes unknowns into research-first rather than scoring them as if known. Two outputs, not one: the ranked table and the 2×2. Recommend; never commit the work. Propose-only.