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/pitiflautico/neobrowser/agents-mdgit clone --depth 1 https://github.com/pitiflautico/neobrowserWrote 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/pitiflautico/neobrowser/agents-md)<a href="https://agentmods.dev/instructions/pitiflautico/neobrowser/agents-md"><img src="https://agentmods.dev/badge/instructions/pitiflautico/neobrowser/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 5.1 | $0.01759 | $0.01759 |
| Opus 5 | $0.00879 | $0.00879 |
| Sonnet 5 | $0.00352 | $0.00352 |
| Haiku 4.5 | $0.00176 | $0.00176 |
Grade A, and why
neobrowser 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 5d 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 — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — orientation for AI agents & contributors
NeoBrowser is an MCP server that drives a real Google Chrome via the Chrome DevTools Protocol (CDP) so AI models can use the web autonomously — stealthily, and optionally with the user's real logged-in sessions.
There are two implementations:
rust/— the shipped product. A single ~5.5 MB binary (dynamically linked against the platform libc, not fully static). All new work goes here.archive/python-oracle/— the original Python implementation, archived. Not the product, not maintained; its README explains why it stopped being a useful oracle.
Build, run, test (Rust — primary)
cd rust
cargo build --release # -> target/release/neobrowser
cargo test # unit + 1 live-Chrome integration test (self-skips w/o Chrome)
cargo fmt --check && cargo clippy --all-targets -- -D warnings # CI gate
./target/release/neobrowser doctor # environment check
./target/release/neobrowser tools --markdown > ../docs/TOOLS.md # regenerate tool docs
python3 scripts/demo.py # end-to-end demo against real sites
node --version # one test syntax-checks the JS embedded in the Rust
Archived Python oracle: cd archive/python-oracle && pip install -e ".[dev]" && python -m pytest -q.
How it works (Rust module map)
| module | role |
|---|---|
cdp |
CDP protocol client on tokio: one connection task multiplexes commands (routed by id) and events; typed timeouts; drains on disconnect |
chrome |
Chrome discovery/launch, stealth flags, health checks; Drop reaps the process |
browser |
Owns the shared Chrome + tabs; lazy launch/attach; self-healing; multi-tab |
capture |
Background listener buffering console + network events per tab |
page |
Page-level CDP verbs (navigate, read, click, type, screenshot, find, nudge_frame) |
ops |
JS-blob tools (fill, form_fill, submit, extract, analyze, scroll, …) |
sessions |
Cookie snapshot/restore, save_session, scripted login |
cookies |
Cross-platform real-profile cookie decryption (Keychain/secret-service/DPAPI) |
reach |
browse/upload/download with an SSRF guard |
search |
Multi-provider search that skips walled sources |
policy |
Central pre-dispatch decision: action class, domain allow/deny, profile (developer/safe/autonomous) |
action |
Verified-action envelope, time budgets, observe → act → verify |
observe |
Accessibility snapshots with stable refs, and diffs between them |
vault |
Encrypted-at-rest session material, TTL, verifiable revocation |
untrusted |
Fencing and labelling page content; injection detection |
trace |
Correlated event timeline, secret redaction, evidence bundles |
devtools |
Web Vitals, computed styles, response bodies, HAR export |
frames |
Shadow DOM / iframe piercing, dialogs, device emulation |
bridge |
Authenticated localhost queue for the Chrome Bridge extension |
config |
Versioned config file, env overrides, named profiles |
walls |
Generic bot-wall / captcha / consent / rate-limit / login detection |
stealth |
JS anti-detection patch (genuine, not spoofed) |
llm |
Optional LLM find fallback (opt-in via ANTHROPIC_API_KEY) |
playbook |
Record/replay tool sequences |
tools / tool_impls / mcp |
Tool trait + registry + argument validation + JSON-RPC server |
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.
- 5d ago First seen · 88 lines · 1,759 tokens per session scan A 4f2548d7fc50
neobrowser AGENTS.md is an instructions file published in the GitHub repository pitiflautico/neobrowser (99 stars, last pushed 9d ago), licensed MIT. It adds 1,759 tokens to every session, about $0.0088 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 2026-08-30.
Other instructions, from other repositories
chrome-local-mcp AGENTS.md
AGENTS.md instructions for thronapple/chrome-local-mcp, covering agents.md, project overview, build & run, register with codex and register with codex cli.
cdpilot CLAUDE.md
Instructions for mehmetnadir/cdpilot, covering cdpilot, kimlik, hızlı başlangıç, dosya haritası and dikkat edilecekler.
cdpilot GEMINI.md
Instructions for mehmetnadir/cdpilot, covering context-mode — mandatory routing rules, blocked commands — do not attempt these, curl / wget — blocked, inline http — blocked and webfetch / web browsing — blocked.
native-devtools-mcp AGENTS.md
AGENTS.md instructions for vectora-foundry/native-devtools-mcp, covering agent context: native-devtools-mcp, 🧠 core reasoning loop, 🗺️ capabilities matrix (strategy guide), 🛠️ tool definitions & schemas and 1. vision & perception (the "eyes").
worbrow AGENTS.md
AGENTS.md instructions for noisystreet/worbrow, covering agents.md, project identity, hard constraints (revert on violation), modifying architecture docs and commit conventions.
Hover CLAUDE.md
Instructions for Hyperyond/Hover, covering directory guide, core documentation index, what hover is, direction (mcp-first, 2026-06) and workspace directories.