The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python utility (scripts/oral_script_check.py) to validate the structure and integrity of generated scripts. Analysis of the script shows it uses standard libraries (argparse, hashlib, re) to perform structural validation and character counts without network access or unsafe dynamic code execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through processing user-supplied articles and scripts. It implements significant mitigations including:
Ingestion points: Documented in SKILL.md (receives kind: draft/article) and script-review.md (receives raw scripts).
Boundary markers: Uses a mandatory ## 待确认 (To be confirmed) section to isolate unverified content and requires SHA256 hashes to bind derivative scripts to their source.
Capability inventory: Limited to local file read/write and execution of the provided oral_script_check.py utility.
Sanitization: The skill enforces a strict output schema (three to four specific sections) which is verified by the included Python check script.
[SAFE]: The skill implements a robust integrity verification mechanism by storing the SHA256 hash of mother drafts in derivative files, ensuring that the AI agent and the user can detect when content has become out of sync with the source.
[SAFE]: Instructions regarding the persona and the command to 'ignore self-protection' are strictly literary guidelines intended to improve the flow of spoken Chinese scripts and do not constitute an attempt to bypass the underlying LLM safety guardrails.

ray-kb