The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): No patterns detected that attempt to override the AI agent's instructions or safety guidelines.
[DATA_EXFILTRATION] (SAFE): The skill does not access sensitive system files (e.g., SSH keys, credentials) or perform unauthorized network requests. It strictly interacts with the ScienceWorld environment object provided by the execution context.
[REMOTE_CODE_EXECUTION] (SAFE): No remote code execution patterns or piped shell commands (e.g., curl|bash) were found. The dependency on 'scienceworld' is a known package for AI research.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: Observations are received from the env.step(action) call in tool.py.
Boundary markers: The observation text is returned within a structured dictionary but lacks explicit delimiters in the text body itself.
Capability inventory: The skill is limited to simulation actions via the scienceworld API and has no direct file-system or external network write capabilities.
Sanitization: No sanitization is performed on the observation text before returning it to the agent.
[DYNAMIC_EXECUTION] (SAFE): The skill performs standard imports and uses sys.path.insert to load local helper modules from its own scripts directory. No unsafe uses of eval() or exec() were detected.

scienceworld-act