The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches arbitrary web content via search.js and content.js and converts it to markdown for the agent to process. This creates a surface for indirect prompt injection where a malicious website could contain instructions designed to influence the agent's behavior.
Ingestion points: content.js (line 46) and search.js (line 92) perform network fetches to retrieve webpage content.
Boundary markers: The skill uses basic separators like --- Result N --- in its output but does not include explicit warnings to the agent to ignore embedded instructions within the fetched markdown.
Capability inventory: The skill can execute shell commands (execFileSync) and perform network operations (fetch).
Sanitization: HTML is processed through JSDOM and @mozilla/readability then converted to Markdown via turndown, which effectively strips executable tags like <script> but preserves natural language instructions.
[COMMAND_EXECUTION]: The script search.js (line 120) uses execFileSync to run the ddgs command-line utility. While it uses the safer array-based argument passing method which avoids shell interpolation, it still relies on an external binary being present and functioning correctly.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external software and libraries to function.
Setup: It instructs the user to install the ddgs tool via uv tool install ddgs and several NPM packages for content parsing and markdown conversion.

web-search