The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from various external sources which are later processed by an LLM for answering questions, creating a surface for indirect prompt injection attacks.
Ingestion points: The +add, +url, and +import commands in SKILL.md are used to add local files, directories, and web content to the knowledge base.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat ingested data as untrusted or to ignore any potential instructions embedded within that data.
Capability inventory: The skill allows for searching and reading from the knowledge base (+grep, +read, +index) and provides an LLM-powered Q&A interface (+ask). It also allows for modification and deletion of knowledge base content.
Sanitization: The instructions do not define any sanitization, filtering, or validation steps for the content retrieved from external sources before it enters the agent's context.
[EXTERNAL_DOWNLOADS]: The +add and +url commands allow for the retrieval of content from arbitrary HTTP(S) URLs, which is then processed and stored within the knowledge base.
[COMMAND_EXECUTION]: The skill utilizes the ae-cli binary to perform operations that interact with the local file system and network, such as reading local ZIP files and uploading content to a remote server.

ae-kb