Browser Workflow

Purpose

Higher-order skill that loads saved browser automation workflows from a directory and executes them through the browser skill. Think of it as a function that takes a workflow file as a parameter, with consistent setup, teardown, and reporting.

Variables

WORKFLOWS_DIR: ./workflows # Directory containing workflow .md files OUTPUT_DIR: ./browser-automations # Where run artifacts are saved

Workflow

1. Parse Arguments

   • First argument → workflow name
   • Scan remaining for keywords (case-insensitive):
     • IF: "headed" → set HEADED=true
     • IF: "headless" → set HEADED=false (explicit override)
     • IF: "vision" → set VISION=true
     • Remaining text → additional prompt to inject into workflow
   • Example: "blog-summarizer https://example.com headed" → workflow=blog-summarizer, HEADED=true, prompt="https://example.com"

2. Validate Environment

   • IF: browser skill is not installed → report "browser skill required but not found" and stop
   • IF: which playwright-cli fails → report "playwright-cli not found. Install it and ensure it's in PATH." and stop
   • IF: WORKFLOWS_DIR does not exist → report "No workflows directory at WORKFLOWS_DIR" and stop
   • Example: missing playwright-cli → "Error: playwright-cli not found. Install it and ensure it's in PATH."

3. Validate Workflow

   • IF: no workflow specified → list available workflows and return
   • Check file exists: <WORKFLOWS_DIR>/<workflow>.md
   • IF: not found → report error with available names
   • Example: no args → "Available workflows: blog-summarizer, price-check"
   • Tool: Glob <WORKFLOWS_DIR>/*.md

4. Load Workflow

   • Read the workflow file
   • IF: file has YAML frontmatter → extract defaults for MODE, VISION
   • Command-line arguments override frontmatter defaults
   • IF: additional prompt set → inject into workflow content
   • Example: workflow says mode: headed but args say "headless" → use headless
   • Tool: Read <WORKFLOWS_DIR>/<workflow>.md

5. Execute Workflow

   • Apply HEADED and VISION settings to the browser skill
   • Create output directory: <OUTPUT_DIR>/<workflow>_<timestamp>/
   • Run the workflow content using the browser skill (open → snapshot → interact → screenshot → close)
   • Save screenshots and artifacts to output directory
   • Example: mkdir -p ./browser-automations/blog-summarizer_2026-03-17_19-30/
   • Tool: Bash

6. Report Results

   • Summarize: workflow name, mode, outcome, output directory, screenshot list, any errors
   • Example:

     WORKFLOW COMPLETE
     Workflow: blog-summarizer
     Mode: headless
     Output: ./browser-automations/blog-summarizer_2026-03-17_19-30/
     Result: Successfully summarized 1 blog post
     

Workflow File Format

Workflow files are markdown in WORKFLOWS_DIR. Optional YAML frontmatter sets defaults that command-line arguments can override:

---
mode: headless
vision: false
---

Navigate to {{ url }}
Click the login button
Fill email with "test@example.com"
Take a screenshot of the dashboard

Family

Part of the browser family. Depends on browser for all browser automation.

• browser — core browser automation (required dependency)
• browser-review — parallel QA orchestration (sibling)
• browser-qa — QA validation agent for individual stories (sibling)
• browser-operator — general browser session agent (sibling)

Works well with

Optional collaborators — browser-workflow runs standalone and these degrade gracefully if absent.

• browser — executes the loaded workflow steps through this underlying capability.