The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources which could contain malicious instructions aimed at hijacking agent behavior.
Ingestion points: The agent is instructed to read the references/cases.md file, which contains numerous URLs to external domains including goodcase.ai and douyin.com.
Capability inventory: The workflow explicitly directs the agent to 'Inspect the anchor's finished media with an available image, video, or browser tool' to analyze the external content at runtime.
Boundary markers: Although the skill requires a 'Preserve/Replace/Avoid' reference contract, there are no technical delimiters or security boundaries to prevent instructions embedded in the fetched media or web pages from being obeyed by the agent.
Sanitization: There is no evidence of sanitization, filtering, or validation of the content retrieved via browser or media inspection tools before it is interpolated into the artifact generation process.
[EXTERNAL_DOWNLOADS]: The skill's operation depends on interacting with external domains that are not part of the standard whitelisted registries.
URL References: The references/cases.md file contains multiple links to goodcase.ai, media.goodcase.ai, and douyin.com used as primary evidence.
Dynamic Fetching: The instructions encourage the agent to perform network operations to fetch and inspect data from these external sites during every invocation.

cinematic-film-still-by-1l42ctj