The Agent Skills Directory

[SAFE]: The skill is entirely composed of documentation, code examples, and analysis procedures for Apple platforms. No executable scripts with malicious intent were found.\n- [PROMPT_INJECTION]: Static detectors flagged keywords such as "ignore previous instructions" and "Disable safety filters" in skills/foundation-models-auditor.md. These are false positives as they are used as search patterns for a security auditor tool to help developers identify vulnerabilities in their own code, not to inject malicious instructions into the agent.\n- [EXTERNAL_DOWNLOADS]: The skill references standard Python packages including coremltools, coreai-torch, and scikit-learn. These are legitimate tools for the Apple machine learning ecosystem and are documented for user-initiated installation from official registries.\n- [INDIRECT_PROMPT_INJECTION]: The auditing utility in skills/foundation-models-auditor.md establishes an attack surface by scanning local project files. However, the procedure is designed for defensive security analysis and explicitly instructs the agent to report vulnerabilities rather than execute them.

axiom-ai