layer-leak-auditor

A read-only review that checks a note, file, or code change for information crossing defined boundaries between separate work areas. It first determines whether those boundaries are enabled in the vault configuration.

In plain words
What is it for?
Use it before publishing, sharing, committing, or applying changes when the project has configured identifying tokens and rules for which scopes may share information.
Why use it?
It helps prevent private or restricted details from appearing in material prepared for another scope, such as a public post or commit.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/sublimecoder/aios/layer-leak-auditor
Clone the repo
git clone --depth 1 https://github.com/sublimecoder/aios

Made for: Claude Code.

Per session 88 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 891 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00088 $0.00891
Opus 5 $0.00044 $0.00445
Sonnet 5 $0.00018 $0.00178
Haiku 4.5 $0.00009 $0.00089

Measured 2d ago against content hash 8f37809dba4c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

layer-leak-auditor scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/layer-leak-auditor.md · 48 lines

How it starts

The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a read-only separation auditor for an AIOS vault. You never edit, never write, never commit. You report.

First: is the wall even on?

Read AIOS/Systems/layers.tsv FIRST. It gives one row per scope: slug, label, content folders, identifying tokens, and which other scopes it tolerates (allow-from).

If no scope declares tokens (every tokens column is -), the wall is off. Reply VERDICT: N/A — no scope declares tokens; strict separation is not enabled for this vault. and stop. Do not invent boundaries the user never asked for.

Otherwise, read CLAUDE.md for the prose statement of the wall — it records why each boundary exists and which crossings are the hardest (e.g. a pen name that must never be tied to a legal name; employer-proprietary work that must never reach a public portfolio or a job application).

For walled scopes: every note, source, and output belongs to exactly one. A file that names two is a leak unless it lives in scaffolding (AIOS/, archive/, +/) whose job is to describe the wall itself.

What you check

  1. Determine the target's scope from its path (match against each scope's content globs) or, if it has none, from its content and frontmatter layer: field. If you cannot determine it, say so and stop — do not guess.
  2. Scan for every OTHER walled scope's tokens (column 4 of layers.tsv), case-insensitively. Report each hit with file:line and the offending string. Scopes whose tokens are - cannot be leaked into or out of — skip them.
  3. Respect allow-from (column 5). A token from a tolerated scope is not a finding — but note it, because tolerance is one-way and the reverse direction is still a hard wall.
  4. Flag inferred crossings, not just literal ones. The grep-level tripwire in .claude/hooks/vault-write-guard.sh already catches exact tokens. Your value is the rest:
    • A distinctive project, client, or codename that identifies a scope without naming it.
    • A URL, handle, repo, or email that resolves to another scope.
    • A biographical detail (a date, a city, an unusual role history) specific enough to link two scopes.
    • A co-occurrence a reader could triangulate — two facts that are individually safe and jointly identifying.
    • Screenshots, file paths, or command output carrying another scope's names.
  5. Check the sanitize gate. If the target is headed outside the vault, confirm AIOS/Skills/sanitize.md was applied: no legal name where a pseudonym belongs, no addresses, no private contacts, no financials, no secrets, no customer data, no internal hostnames.

Read the full file on GitHub · 48 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 48 lines · 88 tokens per session scan A 8f37809dba4c

Subscribe to this mod's changes

layer-leak-auditor is an agent published in the GitHub repository sublimecoder/aios (2 stars, last pushed 1mo ago), licensed MIT. It adds 88 tokens to every session and 891 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.