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 skills/vre/flow-state/chrome-controlnpx skills add vre/flow-state --skill chrome-controlgit clone --depth 1 https://github.com/vre/flow-stateWhat 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 | $0.00029 | $0.00639 |
| Opus 5 | $0.00015 | $0.00319 |
| Sonnet 5 | $0.00006 | $0.00128 |
| Haiku 4.5 | $0.00003 | $0.00064 |
Grade A, and why
chrome-control 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.
How it starts
The opening of the file, as written. The whole thing — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Chrome Control
Control Chrome via CDP. Connects to user's running Chrome — all tabs, cookies, logins accessible.
Prerequisites
User must enable chrome://inspect/#remote-debugging in Chrome.
The daemon starts automatically on first command. No explicit start needed.
User clicks Allow once on the Chrome permission dialog.
Chrome pops a permission dialog on every new WebSocket connection. The daemon holds that single connection open so the user clicks Allow once and all subsequent CLI calls go through without prompts. Each command is a one-shot call; the caller does not hold any connection. The daemon idles out after 5 min of no commands. Unix socket: /tmp/chromectl-{uid}.sock.
If connection fails → tell user to check remote debugging is enabled, or offer legacy mode (launch).
Tool responses are sanitized: prompt-injection markers in page content are stripped before reaching the LLM.
Commands
chromectl.py list
chromectl.py open https://example.com
chromectl.py status
chromectl.py targets
chromectl.py stop
Use --json for machine-parseable JSON output (default is human-readable).
Target commands
All target commands: chromectl.py TARGET COMMAND [ARGS]
Target ID from list/open output. Prefix match OK (e.g. 88FA instead of full 32-char ID).
chromectl.py ABC123 eval "document.title"
chromectl.py ABC123 screenshot -o page.png [--full-page]
chromectl.py ABC123 console-tail [--for 30]
chromectl.py ABC123 navigate https://example.com
DOM helpers
Shorthands for common eval patterns. Selectors are CSS.
click/check/uncheck/highlight SELECTOR
submit/clear FORM
type/fill SELECTOR TEXT
select SELECTOR VALUE
get-text/get-html/get-value/exists/count/get-texts SELECTOR
get-attr SELECTOR ATTR
scroll-to/wait-for/wait-hidden SELECTOR
wait-text SELECTOR TEXT
navigate URL
wait-url PATTERN
scroll-up/scroll-down [PIXELS]
scroll-by X Y
inject-css CSS
reload/back/forward/get-title/get-url/scroll-top/scroll-bottom
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · 89 lines · 29 tokens per session scan A 96639901ffb3
chrome-control is a skill published in the GitHub repository vre/flow-state (12 stars, last pushed 2mo ago), licensed MIT. It adds 29 tokens to every session and 639 once invoked, about $0.0001 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 skills, from other repositories
ask-zilliz
Zilliz Cloud onboarding and usage assistant. Helps users understand Zilliz Cloud, choose the right plan, estimate costs, write code, debug issues, and adopt new features like Functions, Volumes, and Global Clusters. Use this skill whenever the user asks about Zilliz Cloud — including plan selection, pricing, cost…
collection
Use when the user wants to create, list, describe, drop, rename, load, release, or manage collections and collection aliases in Milvus.
diagnose
Use when the user reports that a Zilliz Cloud cluster or Milvus collection is unhealthy, slow, stuck, returning errors, hitting quotas, or otherwise misbehaving — or when they ask "what's wrong with...", "why is ... slow", "diagnose ...", "troubleshoot ...".
cluster
Use when the user wants to create, list, describe, delete, suspend, resume, or modify Zilliz Cloud clusters.
external-collection
Use when the user wants to trigger, describe, or list refresh jobs for an external collection (a collection backed by an external data source such as Vector Lake). Note this is the data-plane refresh workflow, not collection CRUD -- for create/drop/load see the collection skill.
on-demand-cluster
Use when the user wants to create, list, describe, or delete an on-demand (Vector Lake / VectorLake) cluster. On-demand clusters auto-suspend after an idle TTL and are intended for ad-hoc query workloads against a Vector Lake.