---
name: skill-discovery-audit
type: skill
description: >-
  Run a discovery audit at the start of an engagement — assess a venture against
  a chosen category set, capture each observation as a Finding card, cluster the
  findings into Theme Summaries with root cause and How-Might-We questions, and
  tag each by severity. Use this whenever Sonar opens an engagement with an App,
  Design System, or Team audit, or any "where does this fall short of good" review.
  Reads convention-audit and standard-prioritisation; its output feeds skill-prioritise.
  For funnel work, use the sibling skill-conversion-diagnostic. Surfaced by a1.Muse.
---

# discovery-audit

Sonar's opening move. Turns a structured look at a venture into a clustered, severity-tagged set of findings that the rest of the engagement builds on. Produces insight, not a flat list of complaints.

## Trigger

An engagement opens and needs a baseline, or a venture asks "where are we falling short?" Choose this for a quality audit (App / Design System / Team); choose `skill-conversion-diagnostic` when the object is a funnel.

## Behaviour

1. **Scope the object and category set.** Pick the audit object and its category set from `convention-audit`. Confirm what's in and out of scope before looking.
2. **Observe and capture Finding cards.** Work through the categories. Capture each observation as a Finding card (observation, evidence, category, Double-Diamond stage tag, severity per `standard-prioritisation`). Evidence is required — a finding without what-was-seen-where is an opinion.
3. **Cluster into Theme Summaries.** Group findings that share a cause. For each cluster name the *root cause* and write one or more How-Might-We questions. This is where findings become insight; resist one-finding themes unless the finding is genuinely standalone.
4. **Tag severity and accumulation.** Apply `standard-prioritisation` Method A. Check the accumulation rule — where Highs cluster at one step, escalate the step.
5. **Mark confidence.** Tag each theme known / assumed / unknown; route high-severity unknowns to a research-first track rather than asserting them.
6. **Hand off.** The Theme Summaries open Opportunities; pass the set to `skill-prioritise` to score and sequence. Do not jump to interventions inside the audit.

## Reads

`convention-audit` (category sets, card ladder, stage tag), `standard-prioritisation` (severity, confidence overlay), `convention-user-modelling` where the audit touches who the user is.

## Guardrails

Evidence before judgement — every finding cites what was seen. Cluster before prioritising; a flat finding list is not a diagnosis. Stay in Discover/Define — name problems, don't design solutions inside the audit. Mark assumptions as assumptions. Propose-only; the audit informs Aled's decisions, it does not commit work.