probe-traversal

Path traversal protection is tested by attempting to read relative paths targeting sibling, parent, and grandparent directories to verify the platform blocks unauthorized access. The probe confirms that attempts to access surrounding files are handled safely during skill activation.

Skill boundary enforcement secures read access by restricting file operations to the active skill directory, preventing leakage of surrounding data. It validates that the platform's sandbox isolates resources correctly when relative paths are resolved.

Sandbox isolation is verified by running a probe that attempts to read files like ../README.md or ../probe-loading/SKILL.md. The test checks whether the platform successfully blocks or safely handles these cross-directory traversal attempts.

No external dependencies are required to check read access boundaries. The probe operates entirely within the skill's execution environment, validating security rules during activation without needing additional libraries or tools.

Skill activation allowing access to sibling directories indicates a failure in boundary enforcement. Running a traversal probe identifies this by testing restricted read access across sibling paths, confirming whether platform isolation rules are active.

Yes, the path traversal probe provides practical validation for platforms hosting benchmark-skills. It ensures safe isolation of skill resources by confirming that boundary enforcement prevents reading surrounding data during activation.