galiprandi/job-seeker / .agents/skills/browser-automation
browser-automation
VerifiedControl a dedicated browser via agent-browser (Vercel). Use when automating web apps, scraping authenticated sites, filling forms, or navigating SPAs. Do NOT use for desktop apps or API-only integrations.
scripts/README.md
View on GitHub# browser-automation scripts
## `browser.js`
Safe wrapper around `agent-browser` (Vercel CLI). Guarantees repo-scoped isolation: the Chrome profile (`.browser-profile`) and the daemon namespace always resolve to the consuming repo's root, so multiple agents on the same machine never share a browser session. Auto-injects site guides when navigating to a documented site. Runs in-place from this skill directory; never copy it to the consuming repo.
### Wrapper commands
```bash
node .agents/skills/browser-automation/scripts/browser.js open <url> [--headed|--headless]
node .agents/skills/browser-automation/scripts/browser.js goto <url> [--tab <name>]
node .agents/skills/browser-automation/scripts/browser.js tab-new <url> --name <name>
node .agents/skills/browser-automation/scripts/browser.js tab-list
node .agents/skills/browser-automation/scripts/browser.js exec <cmd> [args...] [--tab <name>]
node .agents/skills/browser-automation/scripts/browser.js wait-dom [--timeout <ms>]
node .agents/skills/browser-automation/scripts/browser.js close | close-all
node .agents/skills/browser-automation/scripts/browser.js save-state <path> | load-state <path>
node .agents/skills/browser-automation/scripts/browser.js status
node .agents/skills/browser-automation/scripts/browser.js doctor
node .agents/skills/browser-automation/scripts/browser.js guides
node .agents/skills/browser-automation/scripts/browser.js contribute
```
`exec` forwards everything after it to `agent-browser` (`eval`, `press`, `click`, `fill`, `type`, `keyboard type`, `snapshot`, `find`, `wait`, `screenshot`, `read`, `batch`, ...). `exec goto` is rewritten to `navigate`; `run-code` is rejected (use `eval`/`batch`).
On first run the wrapper creates `agent-browser.json` in the repo root from `agent-browser.template.json` (namespace, profile, headed mode). It never overwrites an existing file.
Environment: `BROWSER_DEBUG=1` for verbose logging; `AGENT_BROWSER_*` vars pass through and override wrapper defaults.
See `SKILL.md` for the full usage reference.
## `transcribe-audio.js`
Local audio→text transcription (sherpa-onnx Parakeet via the `audio-transcription` npm package). No API keys, no cloud — audio never leaves the machine. Auto-installs the package into `~/.cache/agent-tools` on first use; the model/runtime (~640MB) downloads once into `~/.cache/audio-transcription`, shared by all agents on the machine.
```bash
node .agents/skills/browser-automation/scripts/transcribe-audio.js <audio-file> [--out f.txt] [--stdout-only] [--threads N]
```
See `../references/audio-transcription.md` for details and limitations.
## `teams.js`
Helper for sending messages to Microsoft Teams web via keyboard automation. Useful for long or multiline messages where typing line by line manually is impractical.
### Usage
```bash
# Short message (inline)
node scripts/teams.js send <chatId> "short message"
# Long message via stdin (heredoc)
node scripts/teams.js send <chatId> << 'EOF'
📊 Triage report
46 tickets · 27 overdue
[#1234](https://example.com/tickets/1234)
EOF
# Long message from a file
cat /tmp/report.md | node scripts/teams.js send <chatId>
```
### chatId formats
| Type | Format | Example |
|---|---|---|
| Personal notes | `48:notes` | `48:notes` |
| Meeting chat | `19:meeting_xxx@thread.v2` | `19:meeting_NDQ2Nz...@thread.v2` |
| 1:1 chat | `19:*@unq.gbl.spaces` | `19:8:orgid:xxx@unq.gbl.spaces` |
| Group chat | `19:*@thread.v2` | `19:xxx_yyy@thread.v2` |
### Supported Markdown
Teams web parses Markdown in the compose box:
- `[#text](url)` → clickable link with display text
- `**bold**`, `*italic*`, `~~strike~~`
- `- bullet`, `1. numbered`, `> quote`
HTML tags (`<a href>`) are typed as literal text. Use Markdown.
### How it works
All browser calls go through `node browser.js ...` (never `agent-browser` directly):
1. `browser.js goto https://teams.microsoft.com/v2/chat/<chatId>` navigates to the chat
2. `browser.js exec eval` polls for `div[role=textbox]` (up to ~24s; Teams v2 is slow)
3. Focus + clear the textbox (`exec eval` focus, `exec press Meta+a`, `exec press Delete`)
4. Each line is typed with real keystrokes via `exec keyboard type <line>`
5. `exec press Shift+Enter` between lines (Enter alone is a newline, not send)
6. `exec press Meta+Enter` after the last line sends the message
### Requirements
- `browser.js` in the same directory (`scripts/browser.js`)
- Browser logged into Teams (persistent session in `.browser-profile`)
- `agent-browser` installed (`npm install -g agent-browser`) + system Chrome
### When to use it
- Long or multiline messages (reports, triage, summaries)
- Sending from a file or heredoc
For short one-line messages, use the manual commands in the site guide (`sites/teams_com/guide.md`).
## `test-find-guide.js`
Dependency-free test for the site-guide resolution logic in `browser.js` (`findGuide`). Run:
```bash
node .agents/skills/browser-automation/scripts/test-find-guide.js
```
## `browser-share.js` + `viewer.html`
Temporary remote view/control of the repo browser for the user (credential entry, "show me what you're doing", remote login). Default mode restarts this repo's browser with a single tab on the target URL (full isolation; other agents unaffected; cookies persist). `--attach` shares the running session with a remote tab switcher instead.
```bash
node .agents/skills/browser-automation/scripts/browser-share.js <url> [--ttl <sec>] [--viewport <WxH>] [--headed] [--lan]
node .agents/skills/browser-automation/scripts/browser-share.js --attach [--ttl <sec>] [--lan]
```
Enables the agent-browser stream on a random port, serves `viewer.html` (minimal touch-friendly viewer) through a tiny HTTP+WebSocket bridge, and exposes it via an ephemeral `cloudflared` quick tunnel (PATH or `npx -y` fallback) or directly on the LAN with `--lan`. Prints a tokenized link; auto-tears-down after `--ttl` seconds (default and max 300; `--ttl 0` = infinite, explicit only). Requires Node >= 22; cloudflared optional.
The viewer page is a separate `viewer.html` file — the script substitutes `%%K%%` (access token) and `%%EXPIRES%%` (epoch ms) placeholders when serving it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |