The Agent Skills Directory

[SAFE] (SAFE): Analysis of the skill's logic, scripts, and prompts reveals no malicious intent or security vulnerabilities. The implementation follows the principle of least privilege by using only standard library components and local file access.
[Indirect Prompt Injection] (SAFE): The skill processes user-supplied research plans which are later used in prompts. 1. Ingestion points: The design_experiments.py script reads user input from a JSON file via the --plan argument or text via the --method argument. 2. Boundary markers: The generated experiment designs do not use specific delimiters when interpolated into the prompts found in references/stage-prompts.md. 3. Capability inventory: The skill is restricted to local file operations (reading and writing research plans) using the Python standard library and does not have network access or subprocess execution capabilities. 4. Sanitization: Input data is parsed as JSON but not explicitly sanitized for prompt injection markers; however, given the restricted capabilities and the intended use case, this poses no significant security risk.

experiment-design