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/initializ/forge/web-browsenpx skills add initializ/forge --skill web-browsegit clone --depth 1 https://github.com/initializ/forgeWhat 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 | $0.00037 | $0.01099 |
| Opus 5 | $0.00018 | $0.00549 |
| Sonnet 5 | $0.00007 | $0.00220 |
| Haiku 4.5 | $0.00004 | $0.00110 |
Grade A, and why
web-browse 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 — 101 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Web Browse
Drive a real headless browser (Chromium) to accomplish web tasks that a plain
http_request cannot: pages that render content with JavaScript, multi-step
flows, forms, and anything gated behind a click.
The browser is registered only because this skill declares
requires.capabilities: [browser]. All navigation is forced through the
agent's egress proxy, so it obeys the same allowlist, SSRF, and
DNS-rebinding protections as every other web tool. Add the domains you need to
egress_domains; a page that redirects off-allowlist is blocked.
Tools
You drive the browser through six tools. You never see raw HTML — every
observation is a compact digest: the page title, URL, a numbered list of
interactive elements ([3] input(email) "Work email"), and the start of the
page text.
- browser_navigate
{url, wait_ms?}— load a page, get its digest. - browser_state
{max_elements?, scroll_pages?, scroll_to_index?}— re-read the current page (after it changed on its own, to scroll, or to see more elements). Returns a fresh digest with new indices and generation. - browser_click
{index, generation}— click the element atindexfrom the latest digest. Returns the resulting page digest. - browser_fill
{index, text, generation, submit?}— type into an input (or pick a select option by label).submit: truepresses Enter. Password and payment fields are refused unless this skill opts in. - browser_extract
{mode?, selector?, max_chars?, offset?}— pull page content:text(readable markdown, default),links, orhtml(scoped to a CSS selector). Long content is paginated — pass theoffsetfrom the previous call's header to continue. - browser_screenshot
{full_page?, filename?}— capture a PNG for the user (attached to the reply; you will not see the image). Use only when the user wants a visual — read pages with digests and browser_extract.
Workflow
browser_navigateto the target URL. Read the digest's[N]elements and note the Generation number.- To act, pass that
generationtobrowser_click/browser_fillwith the element's index. Every action returns a fresh digest — read it, don't assume the old indices still hold. - If you act on a stale generation (the page navigated or changed), the tool returns an error with a fresh digest — retry using the new indices and generation.
- For long articles or tables,
browser_extractwithmode: textand paginate viaoffset. Usemode: linksto enumerate links.
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 · 101 lines · 37 tokens per session scan A 53107c345db6
web-browse is a skill published in the GitHub repository initializ/forge (156 stars, last pushed 5d ago), licensed Apache-2.0. It adds 37 tokens to every session and 1,099 once invoked, about $0.0002 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
browser-automation
Playwright-based browser automation patterns for autonomous web interaction.
browser-automation
Playwright-based browser automation patterns for autonomous web interaction.
agent-browser
Drive a real headless Chromium on the workspace host to read pages, fill forms, click buttons, run JS, and take screenshots. Use this whenever you need more than a plain HTTP fetch — pages that require JS to render, pages with an interactive element you have to click through, or when you want an accessibility-tree…
browser-automation-cdp
Headless browser automation via Chrome DevTools Protocol (CDP) for dynamic web scraping, UI interaction, and screenshotting.
c-browser
Headless browser automation — navigate pages, click elements, fill forms, take screenshots, scrape content using agent-browser or playwright-cli.
tinyfish-integration
Integrate TinyFish web toolkit (search, fetch, browser automation) into Hermes Agent.