The Agent Skills Directory

[SAFE]: The skill facilitates legitimate knowledge base management tasks through the ae-cli tool. The commands for reading, grepping, and listing sources are standard for document retrieval systems.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection because it allows the ingestion of external data (via URLs and file uploads) that is later processed by an LLM through the +ask command. This is an inherent property of knowledge-base tools, and the instructions include server-side validation steps to manage this risk.
[DATA_EXPOSURE_AND_EXFILTRATION]: While the skill can read and upload files (e.g., +source-read, +source-put), these operations are scoped to the knowledge base management purpose and are gated by user confirmation prompts. No unauthorized network exfiltration of sensitive system files was detected.
[COMMAND_EXECUTION]: The skill uses a fixed CLI interface (ae-cli kb ...) with structured arguments. There are no patterns suggesting arbitrary shell command injection or the execution of unverified scripts.

ae-kb