agentleFS
lohi-ailohi-ai/babysit/.claude/skills/browse/SKILL.md

browse

Skill0 starsChanged 3 days agoMITView on GitHub

Use the browser for focused web-app checks: open a URL, inspect state, click through a flow, capture screenshots, read console errors, or verify a frontend fix. Prefer this over a full QA workflow.

  • Reads credentials
  • Installs packages

58 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use setup-gbrain, open-gstack-browser and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ShreyRaviSkill

    setup-gbrain

    Set up gbrain for this coding agent: install the CLI, initialize a local PGLite or Supabase brain, register MCP, capture per-remote trust policy. One command from zero to "gbrain is running, and this agent can call it." Use when: "setup gbrain", "connect gbrain", "start gbrain", "install gbrain", "configure gbrain for this machine". (gstack)

    Also in 9 projects that use this

  • seek888Skill

    open-gstack-browser

    Launch GStack Browser — AI-controlled Chromium with the sidebar extension baked in. Opens a visible browser window where you can watch every action in real time. The sidebar shows a live activity feed and chat. Anti-bot stealth built in. Use when asked to "open gstack browser", "launch browser", "connect chrome", "open chrome", "real browser", "launch chrome", "side panel", or "control my browser". Voice triggers (speech-to-text aliases): "show me the browser".

    Also in 11 projects that use this

  • Saqlain-811Skill

    ios-sync

    Regenerate the iOS debug bridge against the latest upstream gstack templates. (gstack)

    Also in 6 projects that use this

  • ayudaySkill

    setup-deploy

    Configure deployment settings for /land-and-deploy. Detects your deploy platform (Fly.io, Render, Vercel, Netlify, Heroku, GitHub Actions, custom), production URL, health check endpoints, and deploy status commands. Writes the configuration to CLAUDE.md so all future deploys are automatic. Use when: "setup deploy", "configure deployment", "set up land-and-deploy", "how do I deploy with gstack", "add deploy config".

    Also in 19 projects that use this

  • ShreyRaviSkill

    landing-report

    Read-only queue dashboard for workspace-aware ship. Shows which VERSION slots are currently claimed by open PRs, which sibling Conductor workspaces have WIP work likely to ship soon, and what slot /ship would pick next. No mutations — just a snapshot. Use when asked to "landing report", "what's in the queue", "show me open PRs", or "which version do I claim next". (gstack)

    Also in 9 projects that use this

  • Saqlain-811Skill

    ios-clean

    Remove the DebugBridge SPM package and all #if DEBUG wiring from an iOS app. (gstack)

    Also in 5 projects that use this

  • +3 more from ShreyRavi
  • +1 more from Saqlain-811

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 50 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Prefer the running dev server for local apps.

    line 9
  2. Distinctive:

    Every shell call in this agent session then reuses one window.

    line 42
  3. Very specific:

    If a flow needs a login and the value resolves empty, it's missing from .babysit/.env — report BLOCKED naming that file.

    line 55
  4. Very specific:

    If the run used the iOS provider, record booted simulator UDIDs before the first browser command, shut down only the devices this run …

    line 116
  5. Very specific:

    Never use close --all, pkill/killall on Chrome or Chromium, or simctl shutdown all for routine cleanup.

    line 119
  6. Very specific:

    Do not emit STATUS: DONE* while a browser, daemon, or simulator owned by this run remains.

    line 124

“Common” is measured: kept by at least 6 in 10 of the 50 edited versions of this file we could read. “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

browse

Small browser checks only — a quick smoke test or proof that a touched UI renders. Do not turn this into a product review unless asked.

Rules

  • Prefer the running dev server for local apps; start one only if the task needs it. Capture the smallest useful proof (page state, console errors, screenshot path, one or two interactions) under the ticket or repo temp area. Fix only if the user asked or the calling workflow owns the code change.

Engine

Browsing runs on agent-browser — agent-browser open <url>, then snapshot -i for refs, click @e<n> / type @e<n> <text> / fill / select to interact (the @ref is an argument to a verb, never a bare command), plus eval, screenshot, errors, vitals. One-time: npm install -g agent-browser cloakbrowser. Verbs worth knowing before reaching for --help: batch "<cmd>" "<cmd>" … runs a whole sequence in one call (--bail stops at the first error) — prefer it to one call per click, every call re-reads the whole context; wait <sel|ms>, wait --text <t>, wait --url <pat>, wait --load networkidle; get text|value|url|count <sel>; find role|text|label <value> <action>; press <key>; screenshot [sel] [path] (--full, --if-changed); set viewport <w> <h>, set media dark|light, set offline on|off; console, network requests --filter <pat>; a11y [--selector <css>] (axe-core WCAG audit); diff screenshot --baseline <png>. One agent session = one browser. The session name is the browser identity: the same name reattaches to that browser's daemon (cookies, tabs, window); a different name — or no name, which silently falls through to the shared default session — launches another instance. Never derive the name from the target URL, task, or an ad-hoc label, and never run a bare agent-browser command before these exports:

if [ -n "${CLAUDE_CODE_SESSION_ID:-}" ]; then
  export AGENT_BROWSER_SESSION="cc-${CLAUDE_CODE_SESSION_ID:0:8}"
elif [ -n "${CODEX_SESSION_ID:-}" ]; then
  export AGENT_BROWSER_SESSION="cx-${CODEX_SESSION_ID:0:8}"
else
  export AGENT_BROWSER_SESSION="$(agent-browser session id --scope worktree)"
fi
export AGENT_BROWSER_IDLE_TIMEOUT_MS=900000

Every shell call in this agent session then reuses one window; another agent window gets its own. The idle timeout is a crash backstop for an orphaned daemon, not the normal cleanup path. Refs (@e<n>) are per-session: snapshot the session you're about to act on. If parallel subagents each need their own window, give each export AGENT_BROWSER_NAMESPACE=<agent-id> on top. Sessions are isolated browsers, so login state doesn't carry across agent windows by itself. To share one "profile", add --restore bbs-profile to open (and close): every session loads/saves the same cookies+localStorage bundle under ~/.agent-browser/sessions/, so a login done in one window is there for the next. Don't point concurrent sessions at one --profile dir instead — Chromium locks the user-data-dir per instance. Credentials for a sign-in never live in this skill or the transcript. When a check needs to log in, take them from the project's standard QA env — bbs secrets load exports the gitignored .babysit/.env into the shell, and .babysit/qa.yaml names which vars hold them (standard: QA_USER / QA_PASS):

eval "$(bbs secrets load | sed -E '/^export CLOUDFLARE_(ACCOUNT_ID|API_TOKEN)=/d')"                              # exports .babysit/.env
eval "$(bbs secrets qa probe --env "$(bbs secrets qa default-env)" 2>/dev/null)"
agent-browser type @e<n> "$(printenv "${QA_ENV_USERNAME_ENV:-QA_USER}")"
agent-browser type @e<n> "$(printenv "${QA_ENV_PASSWORD_ENV:-QA_PASS}")"

If a flow needs a login and the value resolves empty, it's missing from .babysit/.env — report BLOCKED naming that file; never hardcode or guess a credential. Default: headed + cloakbrowser stealth. Run agent-browser through cloakbrowser's patched Chromium in a visible window — survives Cloudflare Turnstile / FingerprintJS / DataDome, and a watching human sees the flow. agent-browser spawns the binary itself (it can't attach over CDP), so set these once per shell before any agent-browser call:

export CB="$(npm root -g)/cloakbrowser/dist/index.js"   # cloakbrowser is ESM-only; import by abs path (NODE_PATH won't resolve it)
# pathToFileURL is required: on Windows `npm root -g` returns a backslash path
# (C:\…), which is neither a valid ESM specifier nor survives JS string
# escaping — import() needs a file:// URL. CB travels via env, never embedded
# in the JS source (audit bs-b3m7rnkw #10).
export AGENT_BROWSER_HEADED=1
export AGENT_BROWSER_EXECUTABLE_PATH="$(node --input-type=module -e "import('node:url').then(u=>import(u.pathToFileURL(process.env.CB).href)).then(m=>m.ensureBinary()).then(p=>process.stdout.write(p))" 2>/dev/null)"
# Deterministic --fingerprint seeded from the session name (NOT cwd — a cd between Bash
# calls would change a cwd seed). getDefaultStealthArgs() randomizes --fingerprint on every
# call; that flag feeds agent-browser's per-session launchHash, so a changed value makes it
# tear down and relaunch the browser — a new window, tabs and logins gone.
export AGENT_BROWSER_ARGS="$(node --input-type=module -e "import('node:url').then(u=>import(u.pathToFileURL(process.env.CB).href)).then(m=>{const s=process.env.AGENT_BROWSER_SESSION||'bbs';let h=0;for(const c of s)h=(h*31+c.charCodeAt(0))>>>0;const fp=10000+(h%90000);process.stdout.write(m.getDefaultStealthArgs().map(a=>a.startsWith('--fingerprint=')?'--fingerprint='+fp:a).join(','))})" 2>/dev/null)"

ensureBinary() auto-downloads the ~140MB Chromium on first run (cached at ~/.cloakbrowser/); 2>/dev/null keeps the download banner out of the captured path. All agent-browser commands then work unchanged. Verify with agent-browser eval "navigator.webdriver" → false. Use only for legitimate access checks, never to evade authorization. Each Bash call is a fresh shell — these exports don't survive to the next call. Re-running the blocks per call is safe (session name and fingerprint are deterministic, so the launchHash stays constant and the daemon is reused), but to skip the repeated node startup cost, write all five exports (AGENT_BROWSER_SESSION, AGENT_BROWSER_IDLE_TIMEOUT_MS, and the three stealth vars, plus AGENT_BROWSER_NAMESPACE when set) once to env.sh in the scratchpad and source it at the top of every later call. Parallel subagents share one scratchpad, so each writes its file under a folder named after its namespace (<scratchpad>/<namespace>/env.sh); a shared env.sh lets one agent drive another's window. Don't stash the command in a var hoping word-splitting expands it: zsh doesn't split unquoted vars, so $CMD click @e1 fails — call agent-browser … directly. Drop to plain headless (unset the three stealth vars, or --headless; keep the session and idle-timeout exports) only when there's no display — CI, a remote box without X — or the caller asks.

Browser hygiene

A leaked browser keeps burning CPU and RAM long after the run (130+ orphaned Chrome trees have been found this way), so the lifecycle is part of the check.

  • Own browser for automation: Chrome for Testing, cloakbrowser's Chromium or a Playwright-managed browser, never the human's installed Chrome or its profile. If the installed Chrome is unavoidable, launch it on a cloned --user-data-dir this run owns.
  • Quit, don't kill: end with the browser's own close (agent-browser close, CDP Browser.close) so Chrome shuts down and flushes its profile. kill is a last resort for a process proven owned that survived a scoped close.
  • Check before reaping clones or stale browsers: a clone profile or browser may still be in use (ps for its --user-data-dir, lsof +D <dir>). Reap only what no live process or session owns.
  • Only the exported session: an ad-hoc --session name, or a bare agent-browser call after the exports are lost (fresh shell, compaction), starts a daemon that close on the real session never reaches and that has no idle timeout. Re-source the exports on every call.
  • Whatever else you start that spawns a browser (test suite, render, crawler) is yours to stop through its own shutdown path.

Cleanup

Cleanup is a mandatory finalizer, including on failed checks, tool errors, and BLOCKED exits. After capturing evidence:

  1. Close the exact session with the same session and namespace environment: agent-browser --session "$AGENT_BROWSER_SESSION" close. Then confirm that exact name is absent from agent-browser session list --json. This closes its cloakbrowser/Chrome-for-Testing children and daemon; the idle timeout is only protection for a process that outlives a crashed agent.
  2. If the run used the iOS provider, record booted simulator UDIDs before the first browser command, shut down only the devices this run booted, and quit Simulator.app only when this run launched it and no device remains booted.
  3. Never use close --all, pkill/killall on Chrome or Chromium, or simctl shutdown all for routine cleanup. Those can destroy a human's or sibling agent's work. A stale session from an earlier crash may be closed only when its ownership is proven and its owner is no longer active.

Do not emit STATUS: DONE* while a browser, daemon, or simulator owned by this run remains. Retry its scoped close; if it still cannot be reaped, return BLOCKED with the exact resource and attempted cleanup. Persistent auth state is data, not a live process, and may remain.

Output

End with:

STATUS: DONE | DONE_WITH_CONCERNS | BLOCKED
SUMMARY: <what was checked, what passed, what failed>
EVIDENCE: <url, screenshot/log paths, or "none">
CLEANUP: <closed session and owned auxiliary processes; retained pre-existing resources>

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.