browser
/browser
Browser automation via the `agent-browser` CLI. Use when the user needs to drive websites or Electron apps — navigating, forms, clicks, screenshots, data extraction, testing web apps, UI checks, the dashboard Electron shell, or the user''s own logged-in Chrome via the browser relay or Panerelay (SSO/2FA). Triggers: "open a website", "take a screenshot", "test this web app", "use my own browser".
- Author
- BlackBeltTechnology ↗
- Repository
- pi-agent-dashboard ↗
- Licence
- MIT
- Last repository change
- 25.10.2026
Assessment and security
- Files in this package
- 20
- Scripts
- 2
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 25.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 2 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/check-panerelay.sh, scripts/detect-dashboard.sh
What it does and when to use it
Browser automation via the `agent-browser` CLI. Use when the user needs to drive websites or Electron apps — navigating, forms, clicks, screenshots, data extraction, testing web apps, UI checks, the dashboard Electron shell, or the user''s own logged-in Chrome via the browser relay or Panerelay (SSO/2FA). Triggers: "open a website", "take a screenshot", "test this web app", "use my own browser".
Triggers from the description
Original author description
Browser automation via the `agent-browser` CLI. Use when the user needs to drive websites or Electron apps — navigating, forms, clicks, screenshots, data extraction, testing web apps, UI checks, the dashboard Electron shell, or the user''s own logged-in Chrome via the browser relay or Panerelay (SSO/2FA). Triggers: "open a website", "take a screenshot", "test this web app", "use my own browser".
How it works
- Step 0a — Preflight: agent-browser CLI must be installed
- Step 0b — Auto-detect: route to the right recipe
- Step 1 — Read the matched recipe and execute
- Notes
SKILL.md preview
---
name: browser
description: 'Browser automation via the `agent-browser` CLI. Use when the user needs to drive websites or Electron apps — navigating, forms, clicks, screenshots, data extraction, testing web apps, UI checks, the dashboard Electron shell, or the user''s own logged-in Chrome via the browser relay or Panerelay (SSO/2FA). Triggers: "open a website", "take a screenshot", "test this web app", "use my own browser".'
license: Apache-2.0
allowed-tools: Bash(agent-browser:*), Bash(npx agent-browser:*), Bash(npx @panerelay/setup:*), Bash(curl:*), Bash(pi-dashboard:*)
metadata:
author: pi-dashboard
version: "1.2"
vendoredFrom: agent-browser
vendoredVersion: "0.27.0"
verifiedAgainstCli: "0.33.2"
---
# browser
Composite skill that gives the agent eyes and hands for any browser-driven
task — web pages or Electron desktop apps — via the `agent-browser` CLI.
Three recipes:
- **Web automation** — generic web pages plus Pi Dashboard-specific helpers
(dashboard URL detection, responsive testing, console-error hunting).
Reference: [`references/web.md`](references/web.md).
- **Electron automation** — drive any Chromium-based Electron app, including
a worked example for the Pi Dashboard's own shell via `--debug-cdp`.
Reference: [`references/electron.md`](references/electron.md).
- **The user's own logged-in browser** — reach SSO/2FA-gated sites using the
user's real cookies. Two paths:
- **Pi Dashboard browser** (preferred when the pi session runs on the
dashboard host) — `browser_connect` hands back a loopback CDP URL for the
dedicated profile or a Chrome profile (Playwright Chrome Extension). Reference:
[`references/dashboard-relay.md`](references/dashboard-relay.md).
- **Panerelay extension bridge** (legacy) — a native-messaging host for when
there is no loopback dashboard. Reference:
[`references/own-browser.md`](references/own-browser.md).
## Step 0a — Preflight: `agent-browser` CLI must be installed
The skill does **not** bundle the CLI. Verify it through the dashboard tool
registry — recommend-only, nothing installs without an explicit `--install`:
```bash
pi-dashboard-ensure <extension-package-root>/package.json
```
`<extension-package-root>` is the package that ships this skill — three levels
above this SKILL.md's `.pi/skills/browser/` directory (in a global install:
`$(npm root -g)/@blackbelt-technology/pi-dashboard/node_modules/@blackbelt-technology/pi-dashboard-extension`).
The manifest (`pi.tools`) declares `agent-browser` (resolve) and `chromium`
(pw-browser); the registry answers `present` / `recommended` / `blocked` per
tool.
- `agent-browser: present` → continue to Step 0b.
- `agent-browser: recommended` or `blocked` → halt and tell the user:
> The `agent-browser` CLI is not installed. Install it as a pi extension so
> the `browser` tool is registered in your pi session too:
>
> ```
> pi install npm:pi-agent-browser
> ```
>
> Then re-invoke the skill.
- `chromium` missing → mention the registry's Install hint
(`npx playwright@1.62.1 install chromium`) only when the chosen recipe needs a
CDP browser; the dropdown on the Settings → Tools row carries it.
Do **not** attempt `npm install`, `pi install`, or any other install command
on the user's behalf — they should make that choice explicitly.
## Step 0b — Auto-detect: route to the right recipe
After the preflight, decide which recipe applies. Run this 2-line probe:
```bash
if command -v lsof >/dev/null 2>&1; then
CDP_LIVE=$(lsof -ti :9222 >/dev/null 2>&1 && echo yes || echo no)
else
CDP_LIVE=$(nc -z 127.0.0.1 9222 2>/dev/null && echo yes || echo no)
fi
PD_RUNNING=$(pgrep -f "Pi Dashboard|pi-dashboard" >/dev/null 2>&1 && echo yes || echo no)
# Browser relay: the dashboard's own capability probe (see dashboard-relay.md).
RELAY=$(pi-dashboard api GET /api/browser/status 2>/dev/null || echo unavailable)
echo "CDP_LIVE=$CDP_LIVE PD_RUNNING=$PD_RUNNING RELAY=$RELAY"
```
Routing rule:
| `CDP_LIVE` | `PD_RUNNING` | Route to | Why |
|------------|--------------|---------------------------|---------------------------------------------|
| yes | yes | `references/electron.md` | Pi Dashboard Electron shell is attachable |
| no | yes | `references/electron.md` | Pi Dashboard is up but CDP off — `electron.md` shows the `--debug-cdp` launch instruction |
| any | no | `references/web.md` | No Electron target running; default to web |
**Override**: if the user's request is explicitly about a website (URL,
HTTPS host, "the dashboard at localhost:8000", etc.), route to
`references/web.md` regardless of what's running. Intent wins over capability.
**Override**: if the user's request is explicitly about an Electron app
that isn't Pi Dashboard (Slack, VS Code, Figma, …), route to
`references/electron.md` even when `PD_RUNNING=no` — that recipe covers
launching any Electron app with `--remote-debugging-port`.
**Override (login state)**: if the target needs the user's *existing*
session — "use my browser", "I'm already logged in", SSO/2FA, an internal
tool the agent cannot sign into — route to a login-state recipe regardless of
the probe. Pick by capability, in this order:
1. **`references/dashboard-relay.md`** when the pi session runs on the
dashboard host and the dashboard browser is available. This is the preferred
path; it needs no extra install. Call `browser_connect {}` first (it picks
the dedicated profile by default); without that tool, use the recipe's REST
fallback — the probe's `RELAY` must be a JSON object with `"enabled":true`
and either `dedicated.available` or `"canOpenChrome":true`. A `headless`
result from `browser_connect` for a login task is a stop, not a fallback.
2. **`references/own-browser.md`** (legacy Panerelay bridge) only when the
dashboard browser is unavailable (`RELAY=unavailable`, `enabled:false`) or
the session is remote. `browser_connect` never routes to Panerelay; this
step is yours.
The default bundled browser starts logged out and cannot be made to inherit
those logins; `--profile` copies the profile and still loses macOS
Keychain-encrypted cookies. Prefer the bundled browser whenever login state is
irrelevant — it is faster and more capable. **Never** fall back to it for a
login-state task.
## Step 1 — Read the matched recipe and execute
Read the reference file selected above, then follow its workflow. All
references are self-contained; you do not need to read more than one.
**Challenge encountered**: if at any point the page is an interactive
anti-bot challenge (Cloudflare Turnstile, hCaptcha, reCAPTCHA, "Checking
your browser…") that withholds the content you asked for, stop and follow
[`references/challenge.md`](references/challenge.md). Do not retry, do not
switch `--session`, `--profile`, or provider, and do not run `close --all`.
## Notes
- **Vendoring**: `references/web.md` and `references/electron.md` are
snapshots of upstream `agent-browser` skill content (`core` and
`electron`) at CLI version 0.27.0. See [`UPSTREAM.md`](UPSTREAM.md) for
refresh procedure and [`LICENSE`](LICENSE) for upstream attribution.
`references/own-browser.md`, `references/dashboard-relay.md`, and
`references/challenge.md` are **authored, not vendored** — an upstream
refresh must not overwrite them.
- **Login-state recipes**: `references/dashboard-relay.md` is the preferred
path (Pi Dashboard browser relay); `references/own-browser.md` is the legacy
Panerelay/native-messaging path for remote sessions with no loopback
dashboard.
- **Surfaces share one browser**: the `browser` MCP tool and
`Bash: agent-browser …` drive the *same* daemon and *same* Chrome, with a
shared active page. Probing with one mid-task can silently move the page
out from under the other. Pick one surface per task; use
`--session <name>` for genuinely isolated browsers.
- **Known wrapper bug**: the MCP tool's `eval` echoes the JS source instead
of executing it (and mangles regex escapes) — still present with CLI
0.33.2, so it is a `pi-agent-browser` wrapper defect, not a CLI one. Use
`Bash: agent-browser eval …` instead.
- **No CLI bundled**: agents installing the bridge extension get the
skill text but not the CLI; install on demand per Step 0a.
- **User-local override**: if the user's project has its own
`.pi/skills/browser/` skill, pi's local-wins precedence applies and
this skill is shadowed — that's by design.
Files in this package (20)
- LICENSE 11 KB
- SKILL.md 8 KB
- SKILL.md.AGENTS.md 1 KB
- UPSTREAM.md 3 KB
- references/challenge.md 8 KB
- references/challenge.md.AGENTS.md 1 KB
- references/dashboard-relay.md 18 KB
- references/dashboard-relay.md.AGENTS.md 2 KB
- references/electron.agent.md 4 KB
- references/electron.md 9 KB
- references/electron.md.AGENTS.md 1 KB
- references/own-browser.agent.md 5 KB
- references/own-browser.md 12 KB
- references/own-browser.md.AGENTS.md 1 KB
- references/web.agent.md 24 KB
- references/web.md 81 KB
- references/web.md.AGENTS.md 1 KB
- scripts/check-panerelay.sh 6 KB
- scripts/check-panerelay.sh.AGENTS.md 1 KB
- scripts/detect-dashboard.sh 2 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · BlackBeltTechnology/pi-agent-dashboard | 100 | MIT |
| vibiumdev/vibium | 100 | Apache-2.0 |
| superset-sh/superset | 100 | MIT |
| phnx-labs/agi-cli | 100 | Apache-2.0 |
| phnx-labs/browser-cli | 100 | Apache-2.0 |
| Embervy-ai/embervy | 100 | MIT |
| shibing624/agentica | 100 | Apache-2.0 |
| iammrduncan/fxbot | 100 | Apache-2.0 |
| xarantolus/botler | 100 | MIT |
Database snapshot: 13.11.2026 · package version bb68d5a473b1 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100agent-browser
Browser automation CLI for AI agents; use for web tasks - navigate, click, fill forms, screenshot, scrape, test web/Electron apps. - 100agent-browser
Browser automation via agent-browser CLI. Use when the user needs to navigate websites, fill forms, click buttons, take screenshots, extract data, or test web apps. - 100agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a… - 100agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a… - 100agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a… - 100agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a…