The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest, process, and summarize external research papers (from PDFs or arXiv) and web content. This functionality creates an attack surface where a malicious document could contain instructions intended to override the agent's behavioral guidelines.
Ingestion points: Untrusted data enters the agent context through PDF text extraction and web scraping via tools like paper-figures and claude-defuddle.
Boundary markers: The skill implements mitigations by enforcing a strict section-by-section template (S1-S10) and requiring specific, separate blocks for source provenance (e.g., 'User-supplied', 'From bibliography', and 'Search hits').
Capability inventory: The skill has the capability to read and write files within the local vault, invoke external search/fetch tools, and execute file system commands.
Sanitization: The workflow focuses on summarization and synthesis into markdown templates, and it explicitly instructs the agent to report gaps rather than fabricating content.
[COMMAND_EXECUTION]: The skill utilizes the rg (ripgrep) command-line utility as a fallback mechanism for searching files within the vault's directory structure.
[EXTERNAL_DOWNLOADS]: The skill references and relies on external capabilities and plugins (paper-search, claude-defuddle, paper-figures) to locate and fetch content from research repositories such as arXiv and various web URLs.

ai-wiki