The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown documentation intended for educational and security auditing purposes. It does not contain any executable scripts, binary payloads, or instructions designed to harm the host environment or bypass agent safety guardrails.
[NO_CODE]: There are no scripts, executables, or configuration files bundled with the skill. The analysis confirms the absence of potential execution vectors at the skill level.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and describes vulnerability surfaces where untrusted data (such as poisoned datasets or RAG corpora) can enter an agent's context.
Ingestion points: Identified as third-party model weights, datasets, and RAG corpora referenced in SKILL.md.
Boundary markers: The skill does not define system-level boundaries but discusses the risks of trusting external artifacts.
Capability inventory: No active capabilities or tool executions are defined in the skill itself.
Sanitization: The skill recommends using 'safetensors' as a safe alternative to insecure serialization formats.
[DYNAMIC_EXECUTION]: The documentation highlights the security risks of dynamic execution via unsafe deserialization (e.g., torch.load and pickle.load). It accurately identifies these as RCE vectors in target applications, but does not provide or execute any such payloads itself.

ai-supply-chain