The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data from fact sheets, which could contain instructions aimed at influencing the agent's behavior.
Ingestion points: Content is loaded via facts list and by reading the fact sheet files in the environment to build a mental model of the project.
Boundary markers: The instructions lack specific delimiters to isolate the ingested fact content from the skill's internal logic.
Capability inventory: The skill has access to the facts CLI tool, which allows it to add, edit, or remove entries in the local fact store, as well as run validation checks like facts check and facts lint.
Sanitization: No explicit sanitization or filtering of the fact content is performed before the agent processes the information.
Mitigation: The skill mandates a human-in-the-loop process where the agent must present findings to the user and obtain explicit agreement before applying any changes.

facts-refine