Ref Audit

What it does

When your agent uses it

• Knowledge Management work in your project

Example prompts

• “/ref-audit”

Requirements

• Pre-approved tools (allowed-tools): Read, Write, Bash, Glob, Grep

Workflow steps

3 steps, taken from the step headings in SKILL.md.

1. 1Collect the family + frontmatter
2. 2Run the checks (read-only)
3. 3Write the report + summarise

What it can do on your machine

Read from SKILL.md and the folder at commit fd5e9a4. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves these tools, so the agent can use them without asking each time:

  • Read
  • Write
  • Bash
  • Glob
  • Grep

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Shell commands in SKILL.md call:

  • git

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  No URLs in SKILL.md. Its commands use git, which can reach the network depending on how they are called.

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Ref Audit loads about 1.8k tokens when it runs. Until then it costs about 117 tokens; SKILL.md has 501 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~117

When it runs · the whole SKILL.md, loaded when a task matches

~1.8k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

• NotePre-approves every shell command (allowed-tools: Bash)SKILL.md

  allowed-tools: Read, Write, Bash, Glob, Grep

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from Abilityai/cornelius at commit fd5e9a4, republished under its MIT licence (© Abilityai). 501 words, ~1,783 tokens.

name

ref-audit

description

Integrity report for a reference scope — the reference-scope analog of /coherence-sweep. Scans every entity note for provenance violations (any note not provenance:reference), type/relationship enum violations, missing/invalid as_of, status/superseded_by integrity, dangling wiki-links, orphan entities, and duplicate-entity candidates. READ-ONLY — it reports, it never fixes (fixes are ref-reconcile). Company is the default scope. Safe to run unattended.

allowed-tools

Read, Write, Bash, Glob, Grep

automation

autonomous

user-invocable

true

argument-hint

[scope=Company] [--json]

metadata.version

1.1

metadata.created

2026-07-08

metadata.author

Cornelius

metadata.changelog

1.1: Report template now includes the Freshness (vs SLA) section it already emitted; explicit audit/reconcile boundary note (structural links only…

cd $PROJECT_ROOT
SCOPE="${scope:-Company}"
ls Brain/$SCOPE/*/*.md | wc -l          # total entity notes

cd Brain/Company
# provenance violations
for f in */*.md; do grep -qE '^provenance: reference$' "$f" || echo "PROVENANCE  $f"; done
# type enum violations
grep -rhoE '^type: .*' */*.md | sort -u        # eyeball against the allowed set
# missing as_of / bad date
for f in */*.md; do grep -qE '^as_of: [0-9]{4}-[0-9]{2}-[0-9]{2}$' "$f" || echo "AS_OF       $f"; done
# superseded without a superseded_by
for f in */*.md; do grep -qE '^status: superseded' "$f" && ! grep -qE '^superseded_by:' "$f" && echo "SUPERSEDED  $f"; done
# competitor sitting outside market/
grep -rlE '^relationship: competitor' */*.md | grep -v '^market/' | sed 's/^/PARTITION   /'

# every wiki-link target used
grep -rhoE '\[\[[^]]+\]\]' */*.md | sed -E 's/\[\[([^]|#]+).*/\1/' | sort -u
# every existing entity basename
for f in */*.md; do basename "$f" .md; done | sort -u
# a target in the first list but not the second = dangling

# Ref Audit — Company — YYYY-MM-DD
Total entity notes: N  (people · orgs · products · engagements · market)

## Violations
- Provenance: <count> [list]
- Type/relationship enum: <count> [list]
- Missing/invalid as_of: <count> [list]
- Status/superseded integrity: <count> [list]
- Partition (competitor outside market/): <count> [list]

## Link integrity
- Dangling links: <count> [note → missing target]
- Orphan entities (no relations): <count> [list]

## Duplicate candidates (for ref-reconcile)
- [[A]] ↔ [[B]] — same entity?

## Freshness (vs per-type SLA)
- Over SLA: <count> [note — as_of N days old vs Xd SLA]
- Next cohort to stale: <sub-scope> on <date>

## Verdict: CLEAN | N issues (M require ref-reconcile)

© Abilityai, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/ref-audit of Abilityai/cornelius.

Open the folder on GitHubat commit fd5e9a4