The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch and reference external documentation and prompt templates from a public GitHub repository not managed by a verified or trusted vendor.
Evidence: The file references/cinematic-image-direction.md contains multiple links to github.com/freestylefly/awesome-gpt-image-2 and mandates that the agent "actually read the original text and view the accessible original image" for visual adaptation and case-based learning.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external data from a remote source.
Ingestion points: The agent is directed to ingest Markdown (.md) and JSON files from the freestylefly/awesome-gpt-image-2 repository to guide its image generation and design logic.
Boundary markers: The skill lacks explicit delimiters or instructions to treat the ingested external content as untrusted data or to ignore any embedded instructions within those remote files.
Capability inventory: No evidence of high-risk capabilities such as arbitrary subprocess execution (exec, eval), file system writes, or network exfiltration of sensitive local data was found in the provided files.
Sanitization: There are no mechanisms described or implemented to sanitize, validate, or filter the content retrieved from the external repository before it is used to influence the agent's internal reasoning or output generation.

scene-asset