teacher

SUSPICIOUS. The skill's teaching, artifact-building, and journal features mostly align with its stated purpose, and the flagged `.config` and `eval` patterns look documentation-related rather than malicious. The main security concern is transitive trust: it instructs installation via a skills CLI and operates as a large umbrella skill with many bundled subskills and host integrations, expanding the agent's trusted footprint beyond a simple teaching aid.

SUSPICIOUS. The stated purpose is coherent for video creation, and there is no direct credential theft or exfiltration in this skill, but its trust model is weak: it dynamically installs other skills and may clone and execute a third-party install.sh. The capability footprint is broader than a simple video helper because it bootstraps additional agent instructions from an external repo; that transitive trust and repo-execution path drive the risk.

This is a lesson-code test runner that loads JavaScript, compiles and executes supplied C++, and executes supplied Python to compare outputs against expected results. No malicious intent or data exfiltration is evident, but running untrusted lesson files can execute arbitrary code on the host because vm, g++, and python3 are not security sandboxes. Severity is high when lesson content is untrusted; no suspicious domains, credentials, or environment-variable harvesting are present.