agentleFS
SiteSkillsSiteSkills/siteskills-cowork-plugin/skills/web-browsing/SKILL.md

web-browsing

Skill0 starsChanged 9 months agoMITView on GitHub

Use before any web browsing task — navigating sites, filling forms, clicking buttons, or extracting data. Routes to site-specific skills; universal protocol as fallback.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. shino369/claude-code-personal-workspace★ 4same name
    2. chenwei791129/agent-skills★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository6 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before opening a browser, check whether a skill exists for the target domain:

    line 12
  2. Distinctive:

    Use this priority order (fastest to slowest):

    line 61
  3. Very specific:

    Detect page load without guessing: Use read_page — if it returns content, the page is ready.

    line 95
  4. Distinctive:

    Never silently fail or loop.

    line 116
  5. Very specific:

    Use get_page_text before screenshots for text extraction.

    line 137
  6. Very specific:

    Use find before read_page before screenshots — escalate from fast to slow.

    line 138
  7. Distinctive:

    Never wait "just in case" — check if the page is ready, don't assume it isn't.

    line 139

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Web Browsing — Master Router

Invoke this skill before any browser task. It routes you to a site-specific skill when one is installed, and provides universal fallback navigation for everything else.

Step 1: Check for a Site-Specific Skill

Before opening a browser, check whether a skill exists for the target domain:

DomainSkillCovers
caniuse.comcaniuseBrowser compat data — programmatic path via GitHub raw, shadow DOM gotchas
pub.devpub-devDart/Flutter packages — REST API, search, score, advisories
developer.mozilla.orgmdn-web-docsMDN search API, article navigation, reference lookup
crates.iocrates-ioRust crates — REST API, search, versions, dependencies
pypi.orgpypiPython packages — JSON API, search, release data
news.ycombinator.comhacker-newsHN feed, comments, search via Algolia API

If the target domain is in this table: Invoke the matching skill immediately. Do not open a browser until you have read it.

If the target domain is not in this table: Continue to Step 2.


Step 2: Universal Navigation Protocol

For sites without a specific skill, follow this protocol to navigate efficiently with minimum screenshots.

Reconnaissance (1 screenshot max)

1. Navigate to the URL
2. Take ONE screenshot
3. Use read_page with filter="interactive" to get all clickable/input elements
4. From this single pass, identify:
   - Navigation structure (top bar, sidebar, hamburger)
   - Search capability
   - Auth state (login buttons vs user avatar)
   - Blocking overlays (cookie banners, modals, paywalls)

Clear Obstacles First

Handle these BEFORE your actual task — they block everything:

ObstacleAction
Cookie bannerClick "Reject All" or most privacy-preserving option
Login wallStop — tell the user; never guess credentials
Newsletter popupFind X/close button; if none, press Escape
CAPTCHAStop — tell the user; do not attempt to solve
"Accept terms" modalStop — ask user permission before accepting
PaywallStop — tell the user the content is paywalled

Navigate to Target

Use this priority order (fastest to slowest):

  1. Direct URL — Construct and navigate directly if possible. Zero wasted clicks.
  2. Search — Type query into site's search field, submit, scan results.
  3. Navigation menu — Use read_page or find tool to locate nav links.
  4. Scroll and scan — Last resort. Scroll methodically, screenshot to orient.

Element Finding (fastest to slowest)

TierMethodUse When
1find tool with natural languageButtons, links, labeled elements
2read_page filter="interactive"Full page inventory, forms with many fields
3javascript_tool with CSS selectorKnown selectors, dynamic/hidden elements
4Screenshot + analyze coordinatesLast resort — element unreachable programmatically

Anti-patterns:

  • Taking a screenshot to find a button you could find by label
  • Clicking by coordinate when you have a ref_id
  • Taking multiple screenshots to "make sure" (read_page confirming an element is enough)

Timing and Waiting

SituationWaitDuration
After link click (navigation)wait1–2s
After form submitwait2–3s
After AJAX / dynamic contentwait1–2s
SPA / dashboard initial loadwait3–5s
Between form field fillsnone—
After tab/accordion on same pagenone—

Detect page load without guessing: Use read_page — if it returns content, the page is ready. Check document.readyState === 'complete' via javascript_tool for heavy SPAs.


Data Extraction

Data typeMethod
Article / documentation textget_page_text — order of magnitude faster than screenshots
Tables and listsjavascript_tool → query table/list elements
Visual layout or positioningScreenshot → analyze
Paginated dataExtract page → find "Next" → loop

Error Recovery

  1. Retry once — Click didn't register? Try again. Page didn't load? Wait 2s, check again.
  2. Alternative path — Button not clickable? Try keyboard shortcut. Navigation broken? Try direct URL.
  3. Tell the user — Auth required, CAPTCHA blocking, site down, element truly missing.

Never silently fail or loop. Three attempts max, then escalate to the user.


Common Site Patterns

SPA (React/Vue/Angular): URL changes may not trigger real page loads. Wait for dynamic content, not just document.readyState. Use read_network_requests to detect when API calls finish.

Dashboard/Admin: Left sidebar navigation, breadcrumbs for orientation, settings under gear icon or bottom of sidebar.

E-Commerce: Products at /product/, /p/, or /dp/. Cart at /cart. Search via ?q= or ?search= parameter. Account at /account.

API-backed sites: Try site.com/api/ or check for OpenAPI spec at /openapi.json, /swagger.json, /api/docs. Programmatic API calls are always faster than browser navigation when available.


Performance Rules

  1. One screenshot to orient, then use programmatic tools.
  2. Construct URLs directly when possible — avoids clicking through menus.
  3. Batch form inputs — read all fields once, fill all in sequence, no screenshots between.
  4. Use get_page_text before screenshots for text extraction.
  5. Use find before read_page before screenshots — escalate from fast to slow.
  6. Never wait "just in case" — check if the page is ready, don't assume it isn't.
  7. Tell the user what's happening if something is slow or broken.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.