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 skills add agentproto/ts --skill browsergit clone --depth 1 https://github.com/agentproto/tsWrote 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/skills/agentproto/ts/browser)<a href="https://agentmods.dev/skills/agentproto/ts/browser"><img src="https://agentmods.dev/badge/skills/agentproto/ts/browser/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/agentproto/ts/browser"><img src="https://agentmods.dev/badge/skills/agentproto/ts/browser.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00130 | $0.01493 |
| Opus 5.5 | $0.00052 | $0.00597 |
| Sonnet 5.5 | $0.00026 | $0.00299 |
| Haiku 4.5 | $0.00013 | $0.00149 |
Grade A, and why
browser 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 1mo 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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Browser — the authenticated-browser capability
The browser is an authenticated transport to the user's real sessions. Read data from each site's own internal API (ground truth) and use the DOM only for visuals + the few write buttons. Proven: on a hardened SPA the DOM yields ~0 useful data while an in-page API fetch returns the full structured truth.
Where each concern lives (three homes, one boundary)
- Capability = code. The social capture/search recipes are implemented in
@agstudio/browser-social(theSOCIAL_PLATFORMSadapters +createLinkedInVoyager). Drive the adapter; don't re-derive a recipe by hand. - Operator "how/when" = corpus playbook. The durable judgement a runtime
agent carries (recon SOP, persona = four signals, person-id scoring,
when-to-escalate-stealth, rate-limit + write-gating discipline) is the
social-reconcorpus playbook overlay, injected into the operator's prompt (projects/guilde/packages/builtins/src/knowledge/social-recon.ts; read path inapps/api/src/mastra/guilde-playbook-processors.ts). A runtime operator cannot read these dev skills — the playbook is its home for that knowledge. - Dev-SOP = these skills. What you (building/debugging the stack) need: transport mechanics, the chrome-devtools-mcp tool surface, proven-live gotchas, the per-platform internal-API reference (a superset of what the adapters implement — includes gated write ops). This is the dev-facing mirror.
Golden rules
- API-first for data — find the site's internal API; the DOM is the slow,
obfuscated, anti-scraping surface. News/content sites are the exception →
readability/ld+json (content IS the HTML). See
reference/recon.md. - Fail fast — ~8 s timeouts, not 60 s. After 5-6 s it's ~90 % a hard hang.
- Direct URL nav > clicking navigating-links (clicks that navigate hang).
- Tab/focus discipline —
list_pagesbefore acting; re-assertselect_page({pageId, bringToFront:true})per op under multi-agent contention. - Clear cookie/consent gates — CMP banners + login/paywall cookies first.
- Gate every write — acts as the user; explicit per-action confirmation.
- Capture cheaply — snapshot/evaluate for data; screenshot only for pixels.
- Blocked? → stealth. On a hard anti-bot block ("Sorry, you have been
blocked" / Cloudflare / DataDome / 403-429-503), this Chrome chain won't
recover — switch to the camofox-stealth MCP on agentproto:
mcp_imported_call alias="camofox-stealth" toolName="scrape"(HTTP-first, auto-escalates) orstealth_snapshot. Needs the camofox server on :9377 (launchdcom.agentik.camofox). Seeprojects/browser/apps/stealth-mcp/README.md.
What ships with it
5 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.
- 1mo ago First seen · 106 lines · 130 tokens per session scan A 7448317e5556
browser is a skill published in the GitHub repository agentproto/ts (10 stars, last pushed yesterday), licensed Apache-2.0. It adds 130 tokens to every session and 1,493 once invoked, about $0.0005 per session on Opus 5.5. 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 2026-08-31.
Other skills, from other repositories
demo-recorder
Use when the user wants a scripted screen-recording demo video of a web app, a narration script or text-to-speech voice-over for one, or a dubbed cut.
playwright
Use when the task requires capturing or automating a real browser from the terminal.
playwright-cli
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
browser-tools
Safety rules over the upstream agent-browser skill (scraping ethics, auth, request pacing, local URLs). URL policy is the Claude Code sandbox's job when the operator enables it. Use for browser automation.
qa-screenshot
Screenshot a page (deploy preview or a dev server on localhost) in the sandbox's headless Chromium, look at it, and embed it in a task comment. Use to verify a visual change or record before/after evidence.
skill-creator-primer
You MUST load this skill before the skill-creator skill AND before making ANY change to, or conducting a review of ANY Agent Skill. Triggers include creating, editing, reviewing, or contributing to any part of an Agent Skill (description, frontmatter, body, references, scripts, trigger evals, conflicts, etc).