Fetch web pages and extract structured data using named macros.
npx skills add https://github.com/harmoniis/harmonia --skill browser-harmoniis
Please help me install this Agent Skill. Skill: Browser Source: https://github.com/harmoniis/harmonia/tree/main/lib/tools/browser Command: npx skills add https://github.com/harmoniis/harmonia --skill browser-harmoniis
This Skill provides a secure, self-contained browser tool to fetch web pages and extract structured data using named macros, ensuring no raw HTML reaches the agent and wrapping outputs in a security boundary to prevent prompt injection.
browser_search for single-step fetch+extract and browser_execute for multi-step plans, both returning security-wrapped data.Use browser_search with a URL and macro to fetch and securely extract data.
High-intent search queries and answers about installing and using this skill.
Secure headless browser data extraction uses a 3-layer security model and SSRF protections to fetch web pages safely. It returns structured data through named extraction macros, ensuring no raw HTML reaches the agent and preventing prompt injection.
Available extraction macros for structured web scraping include title, text, links, headings, tables, forms, meta, audio sources, markdown, structured, and smart. These macros enable single-step fetch and extract operations to return formatted data.
SSRF protection when fetching web pages operates through domain allowlists and strict blocking of dangerous endpoints. This controlled fetch mechanism ensures API calls cannot target unauthorized network resources, maintaining a secure boundary.
Rendering dynamic JavaScript pages is supported through Chrome CDP integration, which is available behind a feature flag. When enabled, this option allows the headless browser tool to execute and extract data from dynamically rendered web content.
The difference between single-step and multi-step browser extraction plans is that single-step uses browser_search to fetch and extract in one action, while multi-step uses browser_execute to chain complex extraction sequences, both returning security-wrapped data.
You should avoid automated web data extraction when target domains are not included in the configured allowlist, or when strict SSRF protections block the specific endpoints you need. The security boundary may also limit extraction from sites requiring complex interactions beyond macro capabilities.
charl-krugerControl headless Chrome in Cloudflare Workers via CDP WebSocket.
profmikegreeneParses pilot plans and hypotheses into structured JSON or tabular data for downstream analysis.
dotdevdotdevMap 87 agentwire CLI commands to MCP tool calls for session and pane orchestration.
zhayujieFetch web page HTML and extract title and readable text.