The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as its primary function involves processing untrusted external data.
Ingestion points: Untrusted content is ingested via the WebFetch tool in the retrieve.md, web-scraping.md, and extract-knowledge.md workflows, as well as via yt-dlp transcript extraction in youtube-extraction.md.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within its synthesis prompts.
Capability inventory: The skill possesses significant capabilities, including orchestrating subagents via the Task tool, performing web searches, and executing shell commands.
Sanitization: No content sanitization or validation mechanisms were observed in the workflow definitions.
[COMMAND_EXECUTION]: The skill facilitates the execution of system-level commands and local scripts.
Evidence: The youtube-extraction.md workflow explicitly uses the yt-dlp CLI tool. Additionally, the Bun scripts workflows/claude-research.ts and workflows/perplexity-research.ts import the child_process.exec module, indicating a design that supports shell command execution.

research