The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation instructions fetch and install the asta-plugins component directly from the Allen Institute for AI GitHub repository using the uv package manager.
[COMMAND_EXECUTION]: The skill utilizes bash commands to create experiment directories (mkdir -p) and execute the scientific investigation tool (asta experiment) with user-provided arguments.
[PROMPT_INJECTION]: The skill processes untrusted user data—specifically task descriptions and background knowledge—which are interpolated into shell commands. This constitutes an indirect prompt injection surface as malicious input could influence the agent's execution.
Ingestion points: TASK, BACKGROUND_KNOWLEDGE, TASK_FILE, and BACKGROUND_KNOWLEDGE_FILE variables in SKILL.md.
Boundary markers: Absent; user inputs are directly included in command strings.
Capability inventory: Shell command execution via Bash(asta experiment *) and directory creation via Bash(mkdir -p ...) in SKILL.md.
Sanitization: Absent; no escaping or validation of user-provided strings is specified.

experiment