/canary-automate
Drive a real browser for a one-off task with Canary — navigate, click, fill, scrape, screenshot — and return the result. Nothing is recorded. Use when the user asks to automate a browser task, scrape a page, fill a form, or check something on a site without needing a report.
Claude Code
$ npx -y skills add 0xnyn/canary --skill canary-automate --agent claude-code
Codex
$ npx skills add 0xnyn/canary --skill canary-automate -a codex -y
Cursor
$ npx skills add 0xnyn/canary --skill canary-automate -a cursor -y
Gemini CLI
$ npx skills add 0xnyn/canary --skill canary-automate -a gemini-cli -y
GitHub Copilot
$ npx skills add 0xnyn/canary --skill canary-automate -a github-copilot -y
OpenCode
$ npx skills add 0xnyn/canary --skill canary-automate -a opencode -y
Windsurf
$ npx skills add 0xnyn/canary --skill canary-automate -a windsurf -y
SKILL.md
canary-automate.SKILL.mdname: canary-automate
description: Drive a real browser for a one-off task with Canary — navigate, click, fill, scrape, screenshot — and return the result. Nothing is recorded. Use when the user asks to automate a browser task, scrape a page, fill a form, or check something on a site without needing a report. Trigger phrases — "go to X and get Y", "scrape this page", "automate this browser task", "log in and check", "take a screenshot of".
license: MIT
metadata:
author: usecanary
version: 0.4.4
category: workflow
tags:
- canary
- browser-automation
- scrapingCanary automate (one-off)
Run a script against a real browser and return the result — ephemeral, nothing recorded. Use the **canary-scripting** skill for the API.
When to use
- A quick, one-shot browser task: navigate, extract, fill, screenshot.
- Scraping or checking a page where you don't need a trace / video / report.
- For a recorded, verifiable run **with** a report, use **canary-session** instead.
Examples
Example 1: scrape
User says: "get the top 10 Hacker News titles" or "scrape the headlines" Write a script that opens the page and `evaluate`s the data, run it, return the JSON it logs.
Example 2: check
User says: "is the pricing page up and what's the headline?" or "screenshot the homepage" `goto`, read the element (or `screenshot`), report.
Workflow
1. If the runtime isn't installed: `npx @usecanary/cli install` (one-time; downloads Chromium). 2. Write a short, focused script with the canary-scripting API (`browser.getPage`, `page.goto`, `locator`/`evaluate`, `console.log` the result), observing first on unknown pages (see *Hard rules*). 3. Run it: `npx @usecanary/browser run ./script.js` (or pipe the script via stdin). 4. If the result is empty or a selector missed, **observe and retry**: run a second short script that logs `page.url()`, `page.title()`, and `(await page.snapshotForAI()).full` (or a targeted `locator(...).count()`), pick a better selector, re-run. Named pages persist between runs, so state carries over. 5. Report the script's stdout. On optional extractions, degrade gracefully (log a `WARN`, don't crash) — but don't paper over a miss you can fix by observing. 6. Cleanup (optional): the run leaves a shared background daemon up for reuse. To shut it (and any browser) down, run `npx @usecanary/browser stop` (alias of `canary stop` / `canary daemon stop`).
Hard rules
<!-- canary:snippet rule-observe-first -->
- Unknown page? Snapshot first, then act: read `(await page.snapshotForAI()).full` to see what
is there, pick a semantic selector from it (`getByRole`, `getByText`), then interact. Never guess selectors blind.
- Known page or selectors? Skip the snapshot and use direct selectors — faster and more reliable.
<!-- canary:end rule-observe-first -->
Read more
name: canary-automate
description: Drive a real browser for a one-off task with Canary — navigate, click, fill, scrape, screenshot — and return the result. Nothing is recorded. Use when the user asks to automate a browser task, scrape a page, fill a form, or check something on a site without needing a report. Trigger phrases — "go to X and get Y", "scrape this page", "automate this browser task", "log in and check", "take a screenshot of".
license: MIT
metadata:
author: usecanary
version: 0.4.4
category: workflow
tags:
- canary
- browser-automation
- scrapingCanary automate (one-off)
Run a script against a real browser and return the result — ephemeral, nothing recorded. Use the **canary-scripting** skill for the API.
When to use
- A quick, one-shot browser task: navigate, extract, fill, screenshot.
- Scraping or checking a page where you don't need a trace / video / report.
- For a recorded, verifiable run **with** a report, use **canary-session** instead.
Examples
Example 1: scrape
User says: "get the top 10 Hacker News titles" or "scrape the headlines" Write a script that opens the page and `evaluate`s the data, run it, return the JSON it logs.
Example 2: check
User says: "is the pricing page up and what's the headline?" or "screenshot the homepage" `goto`, read the element (or `screenshot`), report.
Workflow
1. If the runtime isn't installed: `npx @usecanary/cli install` (one-time; downloads Chromium). 2. Write a short, focused script with the canary-scripting API (`browser.getPage`, `page.goto`, `locator`/`evaluate`, `console.log` the result), observing first on unknown pages (see *Hard rules*). 3. Run it: `npx @usecanary/browser run ./script.js` (or pipe the script via stdin). 4. If the result is empty or a selector missed, **observe and retry**: run a second short script that logs `page.url()`, `page.title()`, and `(await page.snapshotForAI()).full` (or a targeted `locator(...).count()`), pick a better selector, re-run. Named pages persist between runs, so state carries over. 5. Report the script's stdout. On optional extractions, degrade gracefully (log a `WARN`, don't crash) — but don't paper over a miss you can fix by observing. 6. Cleanup (optional): the run leaves a shared background daemon up for reuse. To shut it (and any browser) down, run `npx @usecanary/browser stop` (alias of `canary stop` / `canary daemon stop`).
Hard rules
<!-- canary:snippet rule-observe-first -->
- Unknown page? Snapshot first, then act: read `(await page.snapshotForAI()).full` to see what
is there, pick a semantic selector from it (`getByRole`, `getByText`), then interact. Never guess selectors blind.
- Known page or selectors? Skip the snapshot and use direct selectors — faster and more reliable.
<!-- canary:end rule-observe-first -->
QA harness built for Claude Code | E2E testing with screen recordings, console logs, network HARs, and Playwright traces
Other skills on canary.
More skills from other plugins.
api-review
Produce the "Playwright 1.xx API Review" artifact for a range of commits - the public API additions and the breaking or behavior…
promptfoo-evals
Write, refine, run, and QA promptfoo evaluation suites: promptfooconfig.yaml, prompts, providers, vars, tests, assertions,…
deepeval
DeepEval evaluation workflow for AI agents and LLM applications. TRIGGER when the user wants to evaluate or improve an AI agent,…
Questions about canary-automate.
- Which coding agents can use canary-automate?
- canary-automate installs in Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf with the skills CLI, using each agent's flag shown on this page. Flowy's auto-invocation, which fires a skill when your prompt matches, runs in Claude Code only.

