The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted user content, including story segments, file paths, and descriptions of external media references. This creates a surface for indirect prompt injection where instructions hidden in the user-provided data could influence the final prompt output.
Ingestion points: The skill reads user-supplied text, story excerpts, and metadata for images, videos, and audio files provided in the conversation context.
Boundary markers: The skill employs structured templates (e.g., 【生成目标】, 【参考素材职责】) to delimit different parts of the generated prompt, which helps organize output but does not fully sanitize adversarial instructions embedded in user-controlled descriptions.
Capability inventory: The skill has no active capabilities; frontmatter configuration explicitly disables filesystem_read, filesystem_write, tool_use, and network access.
Sanitization: The skill includes explicit instructions to purge sensitive data such as 'interface keys' (接口密钥) and 'running markers' from the final output, reducing the risk of accidental data exposure during the prompt optimization process.

sd25-pe