The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were detected. The skill contains no scripts or executables, relying entirely on natural language instructions for text manipulation.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided transcripts, which represents an ingestion point for untrusted data.
Ingestion points: Transcripts provided as input files or text in SKILL.md.
Boundary markers: The skill enforces a strict 'Zero word addition' (零字新增) rule, which explicitly forbids the agent from generating new text or following instructions not found in the source transcript.
Capability inventory: The skill has no access to shell commands, network tools, or file system writing capabilities.
Sanitization: The instruction set includes a 'Bystander perspective' check to ensure the output remains coherent and focused on the transcript's literal content.
[METADATA_POISONING]: Metadata fields such as the name, version, and description are consistent with the skill's functionality. The author identity and contact information are transparently provided.

xy-clip