The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes capabilities to query and fetch data from external vector database collections, which could contain malicious instructions designed to bypass agent constraints when processed.
Ingestion points: Data is ingested into the agent context through query data and get data commands as defined in references/data.md and references/search.md.
Boundary markers: There are no instructions for the agent to use delimiters or ignore embedded instructions when processing retrieved content.
Capability inventory: The agent can perform significant actions including collection deletion, user management, and backup operations via weaviate-cli.
Sanitization: No sanitization or validation protocols are established for data fetched from the database.
[COMMAND_EXECUTION]: The skill operates by executing shell commands. Specifically, in references/config-management.md, it instructs the agent to source an activation script from a virtual environment located within a user-provided repository path, which could lead to the execution of arbitrary shell code if the path points to a malicious directory.
[EXTERNAL_DOWNLOADS]: The skill guides the user to install the weaviate-cli package from PyPI or via a Homebrew tap, which are external software sources managed by the vendor.
[DYNAMIC_EXECUTION]: The instruction to source <weaviate-cli-repo>/.venv/bin/activate in references/config-management.md involves the runtime execution of a shell script from a path that is not statically defined, representing a dynamic execution pattern.

operating-weaviate-cli