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/captivus/chrome-agent/agents-mdgit clone --depth 1 https://github.com/captivus/chrome-agentWhat 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 | $0.05645 | $0.05645 |
| Opus 5 | $0.02823 | $0.02823 |
| Sonnet 5 | $0.01129 | $0.01129 |
| Haiku 4.5 | $0.00564 | $0.00564 |
Grade A, and why
chrome-agent AGENTS.md scanned grade A with 1 finding 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- **Cookie handoff for bulk.** `Network.getCookies '{"urls":["https://host/"]}'` extracts the session into a `Cookie:` header so a faster external client (`curl`) can fan out a large transfer outside the CDP channel. How it starts
The opening of the file, as written. The whole thing — 255 lines — stays where its author put it; the contents beside it link to each section on GitHub.
chrome-agent
Drive a real Chrome through the Chrome DevTools Protocol (CDP), from the terminal, for AI agents.
Install: uv tool install chrome-agent (or pip install chrome-agent). Requires Google Chrome or Chromium. One runtime dependency (websockets); no Playwright, no browser downloads.
What it is
Address a Chrome instance by name, and send it any CDP command or stream any CDP event. Two channels:
- One-shot (
chrome-agent <inst> Domain.method '{json}') — act: send one command, print the result, disconnect (~70 ms). - Attach (
chrome-agent attach <inst> +Event …) — observe: hold a connection and stream events as JSON lines.
The full protocol, tracked live. chrome-agent forwards your Domain.method straight to Chrome — nothing is validated against a bundled schema. Any command, event, or domain your installed Chrome supports works, including protocol surface newer than this build (e.g. CrashReportContext isn't among the bundled bindings, yet chrome-agent <inst> CrashReportContext.getEntries returns a normal result over the CLI). So help <inst> [Domain[.method]], read live from the browser, is the authoritative, version-correct reference — prefer it over any static list. The typed Python classes are a point-in-time snapshot, not a gate.
experimental ≠ unstable. Most of the live protocol is flagged experimental (domains carry the flag and their members inherit it), and it's tempting to avoid it — don't. Experimental items break at roughly the same rate as the stable core; what predicts churn is how actively a domain is developed, and CDP's busiest domains (Network, Runtime, Page, DOM) are stable-status. The one real experimental signal is removal/rename, and even that is rare. Practical posture: use whatever capability you need regardless of flag; pin the Chrome version you test against; re-verify signatures via help on upgrade.
Operating a page: sense ⇄ act
An agent does two things in a loop: it senses the page and it acts on it. Sensing is the default, continuous mode; acting is the intermittent intervention. After you act you're sensing again — and that perception is both your confirmation of the act and your orientation for the next one. There is no separate "verify" step; the next sense is the verification.
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.
- 2d ago First seen · 255 lines · 5,645 tokens per session scan A 47696f5c82aa
chrome-agent AGENTS.md is an instructions file published in the GitHub repository captivus/chrome-agent (239 stars, last pushed 9d ago), licensed MIT. It adds 5,645 tokens to every session, about $0.0282 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
Browser4 CLAUDE.md
Instructions for platonai/Browser4, covering browser4 — project context for claude, architecture, key dispatch chain (cli → browser), batch commands and e2e test structure.
spring-ai-agentcore AGENTS.md
Instructions for spring-ai-community/spring-ai-agentcore, covering agents.md, project overview, architecture, key components and artifact store classes.
flyto-core CLAUDE.md
Instructions for flytohub/flyto-core, covering claude notes, cross-agent handoff and shared code intelligence.
deckforge AGENTS.md
Instructions for tph-kds/deckforge, covering deckforge agent entry point, code intelligence, read order, default routing and non-negotiable implementation rules.
fast-browser CLAUDE.md
Claude Code instructions for m4ttstack/fast-browser, covering fast browser plugin, where a change belongs, fork branch: use fast-browser-runtime, releasing a new runtime and re-pinning this repo: use the script.
agentoscope AGENTS.md
Instructions for rafaelcg/agentoscope, covering agentoscope — agent instructions, what this is, layout, commands and conventions.