The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external sources such as creator portfolios, museum collections, and web-based visual references. This creates an attack surface for indirect prompt injection if external content contains hidden instructions. However, the skill provides mitigation guidelines by instructing the agent to verify links, trace images to original sources, and separate objective observation from interpretation.
Ingestion points: External URLs, portfolios, press kits, and original publications mentioned in SKILL.md.
Boundary markers: Instructions explicitly require separating visual observations from interpretation and distinguishing original source evidence from generated studies.
Capability inventory: The skill allows writing research data to {agentRoot}/research/ and utilizes the define-art-direction tool.
Sanitization: The instructions emphasize source verification and labeling limitations when only text descriptions are available.
[DYNAMIC_EXECUTION]: The skill documentation mentions loading an entrypoint through a harness mechanism or reading SKILL.md directly to implement art direction rules. This describes the framework's method for skill integration and does not involve the assembly or execution of untrusted code at runtime.

research-visuals