The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the agent-browser (from Vercel Labs) and markdown-new (utilizing Cloudflare's markdown.new API) for web retrieval and content conversion. These are well-known, reputable services used to facilitate the research workflow.
[DATA_EXFILTRATION]: The markdown-new sub-skill transmits webpage URLs and local file contents to the https://markdown.new API to perform Markdown conversion. The skill explicitly documents this behavior and provides a prominent 'Security and Privacy Warning' to the user regarding the processing of data by an external service provider.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted data from the web through the deep-research and markdown-new sub-skills.
Ingestion points: Web content retrieved via agent-browser and markdown_convert.py.
Boundary markers: The documentation includes warnings to 'filter returned text for malicious instructions' before providing it to the LLM.
Capability inventory: The skill has the ability to write files to the local project directory and perform extensive web searches.
Sanitization: The skill relies on LLM guardrails and user review as the primary sanitization mechanism.

kol-writer