The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from third-party Kimi plugins (e.g., S&P Global, Tianyancha, iFinD), creating an indirect prompt injection surface where external content could influence agent behavior.
Ingestion points: Data enters the agent context via screen transcription (screenshots), clipboard access (pbpaste), and reading CSV/Markdown files from local application task directories (e.g., ~/Documents/kimi/tasks/).
Boundary markers: The skill explicitly instructs the agent to use an "Honesty Clause" and "Verification Discipline," requiring it to label data sources and flag missing fields rather than relying on internal model knowledge.
Capability inventory: The agent utilizes GUI automation (computer-use), clipboard interaction (pbcopy/pbpaste), file system access (read/write in specific directories), and execution of local utilities (osascript, ditto, which).
Sanitization: Instructions mandate cross-referencing all extracted values with independent "authority sources" (such as official stock exchange announcements or dedicated CLI tools like qcc) before data is finalized.
[EXTERNAL_DOWNLOADS]: Fetches financial research data from the Oriental Wealth (Eastmoney) official JSON API (reportapi.eastmoney.com). This is a well-known and established financial information service provider.
[COMMAND_EXECUTION]: Executes local system utilities to facilitate automation and file management on macOS, including osascript for window activation, screencapture for visual verification, and ditto for archive extraction.

kimi-use