Chrome CdpSAFE
Give your AI agent access to your live Chrome session — works out of the box, connects to tabs you already have open
Overview
Give your AI agent access to your live Chrome session — works out of the box, connects to tabs you already have open
705e06cd5082OBSERVED · 2026-10-20What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
--- name: chrome-cdp description: Interact with local Chrome browser session (only on explicit user approval after being asked to inspect, debug, or interact with a page open in Chrome) --- # Chrome CDP Lightweight Chrome DevTools Protocol CLI. Connects directly via WebSocket — no Puppeteer, works with 100+ tabs, instant connection. ## Prerequisites - Chrome (or Chromium, Brave, Edge, Vivaldi) with remote debugging enabled: open `chrome://inspect/#remote-debugging` and toggle the switch - Node.js 22+ (uses built-in WebSocket) - If your browser's `DevToolsActivePort` is in a non-standard location, set `CDP_PORT_FILE` to its full path ## Commands All commands use `scripts/cdp.mjs`. The `<target>` is a **unique** targetId prefix from `list`; copy the full prefix shown in the `list` output (for example `6BE827FA`). The CLI rejects ambiguous prefixes. ### List open pages ```bash scripts/cdp.mjs list ``` ### Take a screenshot ```bash scripts/cdp.mjs shot <target> [file] # default: screenshot-<target>.png in runtime dir ``` Captures the **viewport only**. Scroll first with `eval` if you need content below the fold. Output includes the page's DPR and coordinate conversion hint (see **Coordinates** below). ### Accessibility tree snapshot ```bash scripts/cdp.mjs snap <target> ``` ### Evaluate JavaScript ```bash scripts/cdp.mjs eval <target> <expr> ``` > **Watch out:** avoid index-based selection (`querySelectorAll(...)[i]`) across multiple `eval` calls when the DOM can change between them (e.g. after clicking Ignore, card indices shift). Collect all data in one `eval` or use stable selectors. ### Other commands ```bash scripts/cdp.mjs html <target> [selector] # full page or element HTML scripts/cdp.mjs nav <target> <url> # navigate and wait for load scripts/cdp.mjs net <target> # resource timing entries scripts/cdp.mjs click <target> <selector> # click element by CSS selector scripts/cdp.mjs clickxy <target> <x> <y> # click at CSS pixel coords scripts/cdp.mjs type <target> <text> # Input.insertText at current focus; works in cross-origin iframes unlike eval scripts/cdp.mjs loadall <target> <selector> [ms] # click "load more" until gone (default 1500ms between clicks) scripts/cdp.mjs evalraw <target> <method> [json] # raw CDP command passthrough scripts/cdp.mjs open [url] # open new tab (each triggers Allow prompt) scripts/cdp.mjs stop [target] # stop daemon(s) ``` ## Coordinates `shot` saves an image at native resolution: image pixels = CSS pixels × DPR. CDP Input events (`clickxy` etc.) take **CSS pixels**. ``` CSS px = screenshot image px / DPR ``` `shot` prints the DPR for the current page. Typical Retina (DPR=2): divide screenshot coords by 2. ## Tips - Prefer `snap --compact` over `html` for page structure. - Use `type` (not eval) to enter text in cross-origin iframes — `click`/`clickxy` to focus first, then `type`. - Chrome shows an "Allow debugging" modal once per tab on first access. A background daemon keeps the session alive so subsequent commands need no further approval. Daemons auto-exit after 20 minutes of inactivity.
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- declared (2 observation(s))
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (0)
No findings outside the package's declared scope.
Gates applied: no_behavioural_pass.
705e06cd5082full audit observations/trust-audit/skill/pasky__chrome-cdp.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-20 | 705e06cd5082 | SAFE | B | 89 | first audit |
Questions
What does the chrome-cdp — pasky/chrome-cdp-skill skill do?
Give your AI agent access to your live Chrome session — works out of the box, connects to tabs you already have open
Is chrome-cdp — pasky/chrome-cdp-skill safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can chrome-cdp — pasky/chrome-cdp-skill access on my machine?
The audit observed that it reaches the network. Each of those is consistent with what it says it does. Secrets in the source: none found.
How current is this page?
The grade is for one exact copy of the source (705e06cd5082), read on 2026-10-20. The repository is watched, and a new audit runs when it changes — this is the first audit.
Add this badge to your README
Shows this page's audit result and links back to it. It updates when the audit does.
[](https://modelsatlas.com/skills/pasky/chrome-cdp/)
Not your listing, or something out of date? Claim or update your listing.