The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands by embedding user-provided strings (research queries and mission statements) into a JSON payload for the asta generate-theories send-message command. This presents a potential command injection surface if the agent does not properly escape single quotes or JSON delimiters in the user input. Additionally, the inclusion of open * in the allowed tools provides a surface for opening arbitrary local files or external URLs if manipulated.
[PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection by ingesting untrusted research queries and mission statements which are then passed to the Theorizer agent.
Ingestion points: User-provided query, mission statement, and power-user overrides (model_str_primary) in Step 4.
Boundary markers: None explicitly mentioned to separate user input from system instructions.
Capability inventory: The skill can execute shell commands via Bash, access external APIs through asta, and open files/URLs using open.
Sanitization: No explicit sanitization or validation logic is described for the user-provided input before it is used in command construction.
[EXTERNAL_DOWNLOADS]: The skill fetches scientific papers from external literature databases through the asta tool. It also references official documentation and repositories from the Allen Institute for AI (e.g., github.com/allenai/asta-theorizer). These references are to well-known academic and vendor-owned sources and are consistent with the skill's stated purpose.

Generate Theories