The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to invoke multiple external CLI tools, including yt-dlp, perplexity, defuddle, qmd, and obsidian-wiki. These tools are used for data retrieval, transcript extraction, and managing the Obsidian vault index.
[DYNAMIC_EXECUTION]: The skill implements a dynamic command execution engine via the research-backends.md configuration file. It reads YAML-defined command templates (e.g., perplexity search '{query}') and executes them by substituting placeholders at runtime. This allows the skill to run arbitrary binaries with user-defined flags based on local vault configuration.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from WebFetch and WebSearch and interpolates these values into shell command templates using the {url} and {query} placeholders.
Ingestion points: Data enters the system through WebFetch (retrieving web page content) and WebSearch (retrieving search results and URLs).
Boundary markers: No explicit boundary markers or 'ignore' instructions are used when processing the fetched content before synthesis or interpolation.
Capability inventory: The skill possesses the capability to execute shell commands (invoke), write to the local filesystem (Obsidian vault), and perform network operations.
Sanitization: The skill attempts to sanitize inputs by wrapping placeholders in single quotes (e.g., '{url}'), which provides some protection against basic shell injection but may not prevent all forms of argument injection or exploitation of the underlying binaries.

wiki-research