The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local files (CSV, JSON, Excel) and online Feishu/Lark documents, creating a surface for indirect prompt injection. Ingestion occurs in references/source-inspect.md and references/lark-bitable-source.md using the ae-cli and lark-cli tools. While the skill instructs the agent not to print raw values, it lacks explicit content sanitization to prevent the AI from following instructions embedded in the data.\n- [DYNAMIC_EXECUTION]: The skill generates and executes local Node.js, Python, and Shell scripts to facilitate reusable data transformation pipelines as described in references/handoff.md. For example, transform.mjs is generated based on confirmed mappings and subsequently executed using node to process new data files. This behavior is inherent to the skill's primary purpose but represents a dynamic execution pattern.\n- [COMMAND_EXECUTION]: The skill relies on executing various CLI tools and generated scripts to perform its core functions, including ae-cli for Thinking Analytics, lark-cli for Feishu/Lark integration, and local interpreters for generated processing scripts. This is documented across multiple reference files like references/transform.md and references/sink-upload.md.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the @larksuite/cli package from the official NPM registry to enable integration with online documents. This reference to a well-known service is documented in references/lark-bitable-source.md and references/lark-sheet-source.md.

ae-data-integration