The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of ETL pipelines that ingest untrusted content from sources like local files, Amazon S3, and Postgres databases (SKILL.md, flow-patterns.md). This content is processed using LLMs via functions like ExtractByLlm or custom functions.\n
Ingestion points: Data is pulled from LocalFile, AmazonS3, and Postgres sources into the agent's processing context.\n
Boundary markers: There are no patterns or instructions provided for using boundary markers or delimiters to isolate untrusted content from LLM instructions.\n
Capability inventory: The skill enables LLM text generation/extraction, network requests via the requests library, and writing to databases and vector indexes (custom-functions.md, api-operations.md).\n
Sanitization: The documentation does not specify sanitization or escaping mechanisms for input data prior to LLM processing.\n- [COMMAND_EXECUTION]: The documentation includes instructions for executing the cocoindex update --setup -f main command, which utilizes the -f flag to force setup and updates, effectively skipping manual confirmation prompts (SKILL.md).\n- [EXTERNAL_DOWNLOADS]: The skill supports ingesting data from well-known external storage and cloud services including Amazon S3, Azure Blob Storage, and Google Drive, as well as interacting with third-party LLM provider APIs (SKILL.md).

cocoindex