The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external assets (images and text) through tools like asset_analyze, asset_caption, and asset_detect. The output of these analyses (e.g., verdicts or descriptions) is then used by the agent to perform further actions such as tagging, filing into collections, or generating new prompts. This creates a surface where instructions or data extracted from untrusted external assets could influence the agent's behavior.
Ingestion points: Data enters the agent's context through asset_analyze (which accepts up to 10 images and text inputs) and asset_download (used to read analysis results and verdicts).
Boundary markers: There are no explicit delimiters or instructions defined in the skill to treat the output of asset analysis as untrusted data or to ignore embedded instructions.
Capability inventory: The agent has the ability to perform write operations via scenario_tool_execute_write, collection_add_assets, and asset_add_tags based on the results of the analysis.
Sanitization: The skill does not define methods for sanitizing or validating the content extracted from images or text assets before it is used to drive downstream logic.

scenario-asset-analysis