browser-act - Setup Guide¶

Source: browser-act/skills (99,500 installs) Category: Browser Automation Languages: Python + Playwright

Agent-native browser automation framework. Unlike selector-based or accessibility-tree approaches, browser-act uses a "record-and-replay" paradigm - agents record browser interactions as reusable skill templates, then replay them with parameter substitution. Includes browser-act-skill-forge (78.7K installs) for visual template editing.

Installation¶

# Core browser automation
npx skills add browser-act/skills@browser-act

# Skill forge for creating templates (optional)
npx skills add browser-act/skills@browser-act-skill-forge

Verify:

npx skills list | grep browser-act

Prerequisites¶

Requirement	Details
Python	3.10+
Playwright	`pip install playwright && playwright install`
Hermes Agent	Any version
Chrome/Firefox	Installed via Playwright

Install Playwright if missing:

pip install playwright
playwright install chromium

Key Capabilities¶

Record-and-Replay Workflow¶

┌─────────────────────┐     ┌──────────────────────┐     ┌─────────────────────┐
│  Record Interaction │ ──▶ │  Save as Skill       │ ──▶ │  Replay with Params │
│  (browser-act record)│     │  Template (.yaml)    │     │  (browser-act run)  │
└─────────────────────┘     └──────────────────────┘     └─────────────────────┘

Core Commands¶

Command	Purpose	Example
`browser-act record`	Record a new interaction	`browser-act record --name "check-twitter-mentions"`
`browser-act run`	Replay a recorded template	`browser-act run check-twitter-mentions --account CorpusIQ`
`browser-act forge`	Open visual template editor	`browser-act forge check-twitter-mentions`
`browser-act list`	List saved templates	`browser-act list --category social`
`browser-act params`	Show template parameters	`browser-act params check-twitter-mentions`

Parameter Substitution¶

Record once, replay with different values:

# Template: check-twitter-mentions.yaml
steps:
  - action: navigate
    url: "https://twitter.com/{{ account }}"
  - action: click
    selector: "[data-testid='mentions-tab']"
  - action: extract
    selector: "[data-testid='tweet']"
    fields:
      - author: "[data-testid='User-Name']"
      - text: "[data-testid='tweetText']"
      - time: "time"

# Replay for different accounts:
browser-act run check-twitter-mentions --account CorpusIQ
browser-act run check-twitter-mentions --account nousresearch

Anti-Detection Features¶

Feature	Purpose	Config
Fingerprint rotation	Randomize browser fingerprint per session	`--fingerprint random`
Human-like timing	Random delays between actions (150-800ms)	`--human-timing true`
Proxy rotation	Route through different IPs	`--proxy pool:residential`
CAPTCHA handling	Auto-detect and pause for manual solve	`--captcha pause`
Session isolation	Separate browser profiles per template	Default; configurable

Skill Forge (Visual Editor)¶

browser-act forge check-twitter-mentions

The visual editor lets you: 1. View recorded steps as a flowchart 2. Edit selectors with live element highlighting 3. Add assertions (element visible, text contains, URL matches) 4. Test parameter substitution with sample values 5. Debug failures with step-by-step replay and screenshots

Common Templates for CorpusIQ¶

# Record: Login → check notifications → extract mentions → reply
browser-act record --name "social-check" --start-url "https://twitter.com/notifications"

Competitive Research¶

# Record: Visit competitor page → screenshot pricing → extract features
browser-act record --name "competitor-analysis" --params "url,company_name"

Platform Health Check¶

# Record: Login to Postiz → check scheduled posts → verify publishing status
browser-act record --name "postiz-health" --start-url "https://app.postiz.com"

Comparison: browser-act vs agent-browser¶

Feature	browser-act	agent-browser
Interaction model	Record-and-replay	Accessibility tree (refs)
Best for	Repeated workflows, social media	One-off exploration, scraping
Template reuse	Yes - parameter substitution	No - manual each time
Anti-detection	Built-in (fingerprint, timing, proxy)	Minimal (Chrome for Testing)
Learning curve	Moderate (record once)	Low (snap + ref)
Speed	Slower (human timing)	Fast (native Rust)
Install base	99.5K	553.6K
CorpusIQ use case	Daily social checks, competitor monitoring	Web scraping, data extraction

Recommendation: Use both. browser-act for recurring workflows (daily social monitoring, competitive checks). agent-browser for ad-hoc research and scraping.

Troubleshooting¶

Symptom	Cause	Fix
`browser-act: command not found`	Skill not installed	`npx skills add browser-act/skills@browser-act`
Template replay fails on login	Session expired	Re-record with fresh login, save cookies
CAPTCHA blocks recording	Anti-bot detection	Enable `--human-timing` and `--fingerprint random`
Selector not found on replay	Site structure changed	Use `browser-act forge` to update selectors
Recorded template too brittle	Overly specific selectors	Use `--robust-selectors` flag when recording

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects