The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a specific local shell command: node animation-method/scripts/compile-animation-ast.mjs episodes/<slug>/09-导演. This script is used to deterministically generate the animation-ast.json file from approved briefs.
[DYNAMIC_EXECUTION]: The production workflow relies on runtime generation of metadata through the execution of the compile-animation-ast.mjs script. The skill enforces that the resulting file must be verified with a checksum (--check) to ensure it matches the approved input.
[INDIRECT_PROMPT_INJECTION]: The skill processes external text data from sources such as 01-口播稿.md and 07-剪辑/subtitles.srt which are susceptible to instruction injection.
Ingestion points: Untrusted data enters the context through local markdown and subtitle (SRT) files mentioned in the 输入 (Input) section of SKILL.md.
Boundary markers: There are no explicit instructions to the agent to treat the contents of the script or subtitles as data only or to ignore embedded instructions.
Capability inventory: The skill has the capability to write multiple files to the 09-导演/ directory and execute local scripts via Node.js (SKILL.md).
Sanitization: The skill includes integrity checks using SHA-256 hashes defined in edit-manifest.json and shooting-record.json, and it requires a variance review (spoken-script-variance.json) to detect deviations, but it does not specify sanitization for the actual content processed by the agent.

laohan-daoyan