markdown-browser

What This Skill Does

The markdown-browser skill acts as an essential orchestration layer for OpenClaw agents, designed specifically to sanitize, normalize, and enforce policy compliance on web-fetched data. Rather than replacing the foundational web_fetch functionality, this skill serves as a secondary processor that consumes the raw output of web_fetch to ensure that downstream agent logic receives clean, structured, and privacy-compliant data.

The skill operates via a multi-layer MECE (Mutually Exclusive, Collectively Exhaustive) architecture. First, it processes content signals to determine policy actions such as allow_input or block_input, ensuring the agent respects data usage policies. Second, it implements a privacy layer that scrubs sensitive query parameters and URL fragments, maintaining debuggability while protecting user security. Finally, the normalization layer handles content conversion; it preserves native Markdown, provides a turndown-based fallback for HTML, and handles miscellaneous text types, delivering a unified output schema regardless of the source page format.

Installation

To integrate this skill into your OpenClaw environment, use the following CLI command within your terminal:

clawhub install openclaw/skills/skills/2233admin/markdown-browser

Ensure that you have the runtime dependencies installed by navigating to the skill directory and running npm install --omit=dev. This prepares the environment for high-performance processing of web results.

Use Cases

• Automated Research Agents: Standardizing diverse web content (HTML/text/Markdown) into a single format for reliable LLM ingestion.
• Content Policy Enforcement: Automatically blocking inputs from sensitive or disallowed domains based on Content-Signal headers.
• Privacy-First Data Logging: Removing sensitive path/query data from internal logs while keeping the root URL for debugging purposes.
• Reliable Token Management: Utilizing x-markdown-tokens to accurately estimate the context window usage before passing data to the LLM.

Example Prompts

1. "Fetch the documentation from this URL and use the markdown-browser skill to ensure the content is safely formatted for my knowledge base."
2. "Get the latest news from this tech blog, sanitize the URL privacy information, and summarize the article using the standard normalized format."
3. "Run a web fetch on this research page, check the content signals, and tell me if the policy allows me to process this data for training purposes."

Tips & Limitations

• Pre-fetch Dependency: Always ensure your agent has executed an official web_fetch prior to invoking this skill; it is strictly a post-processing tool.
• Normalization Fallback: While the turndown converter is highly robust for HTML-to-Markdown tasks, keep in mind that complex JavaScript-heavy sites may lose some interactive elements during normalization.
• Token Estimation: The token_estimate field is an approximation. For highly critical memory-constrained tasks, allow for a 10% overhead buffer.
• Policy Transparency: If an action is marked as block_input or needs_review, the agent should yield control to the human user rather than attempting to bypass the signal.