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.
npx agentmods add instructions/dylansantwani/openbrowser/agents-mdgit clone --depth 1 https://github.com/dylansantwani/openbrowserWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/dylansantwani/openbrowser/agents-md)<a href="https://agentmods.dev/instructions/dylansantwani/openbrowser/agents-md"><img src="https://agentmods.dev/badge/instructions/dylansantwani/openbrowser/agents-md.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00884 | $0.00884 |
| Opus 5 | $0.00442 | $0.00442 |
| Sonnet 5 | $0.00177 | $0.00177 |
| Haiku 4.5 | $0.00088 | $0.00088 |
Grade A, and why
openbrowser AGENTS.md 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 4d 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.
How it starts
The opening of the file, as written. The whole thing — 73 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
Guidance for AI coding agents working in this repo. CLAUDE.md is the long
version with the war stories behind each rule; read it before non-trivial work.
What this is
Two halves, two languages of failure:
mcp-server/— Node. A zero-dependency MCP stdio server + hub. Bugs here look like "the tool never appears in my client" or "every call times out".extension/— Chrome Manifest V3. Bugs here look like "it says it worked but nothing happened".
Build, test, run
- No build step, no
npm install— the repo has zero dependencies. npm test— 232 tests, no browser needed. Run before and after every change. Run it a few times when touchingws.jsorhub.js(some races are intermittent).npm run preview— serves the UI and the in-browser accessibility-tree tests athttp://localhost:8850/test/a11y-browser.html(needs a real viewport).npm run hub— run the hub standalone, verbose.node mcp-server/src/index.js --health— check whether Chrome is connected.
Hard rules
- Never add a dependency. If you need a library, write the ~200 lines instead. This is the project's most important property.
stdoutis protocol-only inmcp-server/. Non-JSON-RPC output corrupts the stream; human-readable logging goes to stderr.- The service worker has no durable memory (MV3 kills it after ~30s idle).
Durable state lives in
chrome.storage; every top-level statement inbackground/must be safe to run many times. - Write errors for a model to recover from. Put the recovery action in the
message:
"ref e12 no longer exists — take a fresh snapshot", not"Error: null". - Sessions must not touch each other's tabs or windows. Anything keyed by an
agent-chosen name is keyed by
(session, name). Never act on a tab the session did not choose;_sessionand_browserare stamped server-side after the args spread so a model cannot spoof them. element.click()is ignored by serious sites (isTrusted: false). All pointer/keyboard input goes through CDP's Input domain.
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.
- 4d ago First seen · 73 lines · 884 tokens per session scan A 991704012fc1
openbrowser AGENTS.md is an instructions file published in the GitHub repository dylansantwani/openbrowser (1 stars, last pushed 3d ago), licensed MIT. It adds 884 tokens to every session, about $0.0044 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.
Other instructions, from other repositories
chromeboost GEMINI.md
Gemini CLI instructions for lordamdal/chromeboost, covering chromeboost — gemini instructions, what chromeboost is, when to use chromeboost (be proactive), hard rules — never break these and standard flow pattern.
browser-bridge AGENTS.md
Instructions for koltyakov/browser-bridge, covering project guide, data flow, package map, where to find things and method dispatch (extension side).
browser-mcp CLAUDE.md
Claude Code instructions for drisplabs/browser-mcp, covering claude.md, project overview, build and development commands, build and quality checks.
browser-agent AGENTS.md
Instructions for uiuing/browser-agent, covering agents.md — guide for ai coding agents, setup & commands, tests — run these before claiming done, repository layout and architecture invariants (do not break these).
browser-mcp AGENTS.md
AGENTS.md instructions for drisplabs/browser-mcp, covering snapshot & interaction pipeline — audit results, critical (agent can't complete tasks), major (agent works around it poorly), m2. batch vs. fallback coordinate space mismatch and m3. wrapping hidden inputs dropped from snapshot.
nekoro-browser AGENTS.md
Instructions for zeshuochen/nekoro-browser, covering code priorities, testing and contributing.