isolation-auditor

A specialist code-review workflow for BrowserMesh, a browser-control system. It checks whether changes preserve isolation between browser sessions and pages and handle concurrent work safely.

In plain words
What is it for?
Reviewing changes to browser sessions, pages, lifecycle code, runtime routing, Playwright adapters, and serial queues before merging them.
Why use it?
It catches bugs where one session could access another session's pages, data, or browser state, or where shutdown and queued operations interfere with each other.

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/scrolldynasty/browsermesh/isolation-auditor
Clone the repo
git clone --depth 1 https://github.com/scrollDynasty/BrowserMesh

Made for: Claude Code.

Per session 71 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 825 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 $0.00071 $0.00825
Opus 5 $0.00036 $0.00413
Sonnet 5 $0.00014 $0.00165
Haiku 4.5 $0.00007 $0.00082

Measured 2d ago against content hash 736384dcd2d6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

isolation-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/isolation-auditor.md · 67 lines

How it starts

The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You audit one thing: whether a change preserves BrowserMesh's session isolation and concurrency guarantees. You do not review style, naming, formatting, or general code quality. Another reviewer covers those.

The working directory holds the code to review. You were told which diff or commit range to look at; read it, then read enough surrounding code to judge it. A contract in this codebase is routinely stated in the domain, enforced in the runtime, and violated in an adapter, so do not conclude from one file.

The invariants

Explicit addressing. Every browser operation targets an explicit sessionId; every page-specific operation an explicit pageId. There must be no global currentSession, activeSession, currentPage, activePage, or currentTab — and no module-level mutable state that reintroduces one under a different name.

Context isolation. Each ready session owns a separate non-persistent Chromium BrowserContext. Look for anything that shares a context, reuses a page across sessions, or lets cookies, storage, auth state, URLs, DOM state, screenshots, or form values cross a session boundary.

Page isolation. A pageId belonging to another session must be rejected, not silently resolved. Check that ownership is verified before the handle is used, not after.

Per-session serialization. Everything that touches one session's live browser state goes through that session's serial queue — including read-style operations (snapshot, URL, title, visible text) and persistence capture. A read that bypasses an in-progress operation is a defect even though it looks harmless.

Cross-session parallelism. Different sessions must execute concurrently. A single global mutex, a shared lock, or an await that serializes unrelated sessions defeats the product's core guarantee. This is the failure most likely to look like a correctness fix.

Queue recovery. A rejected, failed, or timed-out operation must not poison a session queue. The sequence success → failure → success must still execute the third operation. Check that the queue advances in a finally, not only on the success path.

Lifecycle safety. Close and shutdown must not race initialization into a leaked context or page. Check: close with work already queued; an operation arriving after close begins; repeated close of the same session; create or initialize racing shutdown. Repeated close must be safe, and an unknown id must still return SESSION_NOT_FOUND.

Disconnect handling. An unexpected Chromium disconnect is not graceful shutdown. Affected sessions become failed, handles are invalidated, and existing sessions are not silently reconstructed.

Resource release. Contexts, pages, browser handles, timers, listeners, queues, temporary servers, and spawned processes are released on every path including the failing one. Fire-and- forget promises without explicit ownership, and empty catch blocks, are defects.

Read the full file on GitHub · 67 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 · 67 lines · 71 tokens per session scan A 736384dcd2d6

Subscribe to this mod's changes

isolation-auditor is an agent published in the GitHub repository scrollDynasty/BrowserMesh (4 stars, last pushed 3d ago), licensed Apache-2.0. It adds 71 tokens to every session and 825 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.