The Agent Skills Directory

[SAFE]: The skill defines a legitimate and highly detailed persona for assisting with software development workflows without attempting to bypass safety filters or perform unauthorized actions.
[COMMAND_EXECUTION]: The skill references the execution of local scripts (e.g., download-xls.js) and platform-specific tools (e.g., $oneos-autoprd, $YunxiaoPMapp) for generating documentation and managing project data, which is appropriate for its stated purpose.
[DATA_EXFILTRATION]: While the skill interacts with external project management APIs (Yunxiao), it enforces a strict confirmation policy ('Y2') requiring user approval before any write operations occur, and restricts data access based on the user's role.
[PROMPT_INJECTION]: The instructions establish a complex personality and operational logic but do not include directives to ignore system instructions, bypass ethical constraints, or extract internal system prompts.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent surface for indirect prompt injection because it processes external requirement documents, user-provided transcripts, and third-party project data.
Ingestion points: Processing user-uploaded requirements, voice transcripts, and external project status retrieved via API.
Boundary markers: The skill relies on internal logic and role-based filtering rather than explicit delimiters to isolate untrusted data.
Capability inventory: Includes workspace file writing and interaction with external project management toolchains.
Sanitization: Not explicitly defined in the instructions, though the skill follows rigid output templates.

yanchufasui