audit-brain — score an existing knowledge base

This skill is an ordered checklist: each step opens when the one before it has produced its result. It reports first and changes nothing until the user has decided. Scaffolding a vault from scratch is init-brain, not this skill.

Whenever a step puts more than one decision to the user, it asks them in the round shape the grill-point skill gives in its step 3.

The component reference is the brain-anatomy skill — read it before step 2.

Steps

1. Discover the project's conventions — read the host project's agent instructions file (AGENTS.md, or CLAUDE.md) and inspect the vault's own shape: which trees exist, where the constitution lives, and the language the vault writes its documents in. A rule the vault states about itself governs the anatomy's default, and the scorecard is written in the vault's language. Then show the user the plan for this run — its steps by number with a 🛑 on each one that stops for the user, then one line of what this step found and one of what it could not find, labelled in the user's language — and wait for the user's go-ahead, or take the dispatch as the go-ahead when this run was sent to work without the user, as a sub-agent; name each later step by its number as it starts.

2. Score all 12 components — walk the brain-anatomy skill against the existing vault and mark each component present / partial / missing, citing a file path (or its absence) as evidence. Every component accounted for — no skips. Where an earlier scorecard exists, mark each component it did not score: those rows are what the skill set has added since the last audit, which makes the scorecard the vault's own record of what an update changed. The scorecard is one table, one row per component in the anatomy's order, headers in the vault's language — the same columns every audit, so one scorecard can be read against the last:

   #	Component	State	Evidence	Outcome
   3	Indexes at every level	partial	index.md lists 3 of 5 subjects

   State is present, partial, or missing; Evidence is the path, or the path that was looked for and is absent; Outcome stays empty until step 4 fills it. A component the earlier scorecard did not score opens its Evidence with "new since" and that scorecard's date, so the mark adds no column.

3. Record the scorecard, then settle the fixes with the user — write the scorecard to a dated file in the vault (default docs/brain-audit-<YYYY-MM-DD>.md; follow the vault's own docs convention if it has one). This step is a grill point: run the grill-point skill, and build nothing on a decision the grill has not settled. Each gap is one decision, with a concrete proposal as its recommendation — how many trees, where a kind of file lands, what is archived — because those are choices only someone who works in the vault can judge. Change nothing in the vault until the grill has ended.
4. Apply the decided fixes and verify — implement only what was decided, and run the project's check script where one exists, then confirm: the scorecard carries the line its grill wrote, every decided fix is present in the vault, and no link broken by it. Close the loop in the scorecard file: mark each gap's Outcome as fixed or declined, so the next audit starts from what was already decided rather than re-proposing it. Last, delete this skill's grill log, GRILL-LOG-<skill>.md, from the run's folder: a grill log left behind is a second copy of answers that drifts from the artifact, against which the next run would check its drafts.