The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process a wide variety of untrusted document formats (Word, PDF, Excel, PowerPoint, and photos). These files serve as external data sources that could contain embedded adversarial instructions intended to influence the agent's behavior during processing.
Ingestion points: The SKILL.md identifies multiple untrusted inputs: DOCX, PDF, PPTX, Excel (.xlsm), photos, and scans.
Boundary markers: There are no explicit instructions for the agent to use delimiters or sanitization protocols when reading content from these documents.
Capability inventory: The router triggers the loading of sibling skills (e.g., doc-to-markdown, pdf-creator, excel-automation) which possess the capabilities to read, write, and convert files.
Sanitization: No sanitization or filtering of external content is specified in the routing logic.
[DYNAMIC_EXECUTION]: The skill requires the agent to dynamically calculate and resolve paths to load sibling instruction sets (SKILL.md files). It instructs the agent to resolve canonical locations, follow symlinks, and identify a "suite root" at runtime to locate and load dependencies that are intentionally hidden from standard discovery tools.
[TIME_DELAYED_CONDITIONAL]: The instructions include a conditional trigger that automatically routes the agent to the docs-cleaner skill following authorized code or configuration changes. This behavior is designed to occur even if the user has not explicitly requested a documentation check or cleanup, representing an automated action based on environmental state.

docs-router