The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands using the renku CLI (renku generation models list/import/show/remove) to manage personal model libraries. Additionally, the skill runs pnpm test within a source repository to validate authored changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data, including model apiId, name, and research notes, to update discovery files and Markdown guides. This presents an attack surface where malicious instructions could be embedded in the model metadata.
Ingestion points: User requests for adding/refreshing models, as well as external research sources and provider documentation.
Boundary markers: No explicit delimiters or warnings are mentioned to sanitize or isolate user-provided metadata before it is written to files.
Capability inventory: The skill possesses significant capabilities, including executing shell commands (renku, pnpm) and writing to JSON and Markdown files in a local repository.
Sanitization: There is no description of sanitizing or validating the content of user-provided strings before they are interpolated into configuration files or passed as arguments to the renku CLI.

model-researcher