The Agent Skills Directory

[SAFE]: The skill provides security-hardened guidance for other AI agent skills. It explicitly requires redacting secrets (API keys, tokens, credentials) from all logs, previews, and reports (found in references/dependency-install.md and references/conversion-workflow.md).
[SAFE]: Includes defensive instructions against prompt injection by requiring agents to treat all user-supplied content (code, comments, requirements files) as evidence to inspect rather than instructions to obey (references/conversion-common.md).
[SAFE]: Enforces secure handling of external dependencies by requiring static inspection, auditing for typosquats or malicious URLs, and mandatory user confirmation before any installation (references/dependency-install.md).
[SAFE]: Promotes safe deserialization practices by mandating weights_only=True when loading PyTorch checkpoints using torch.load (references/conversion-workflow.md).
[SAFE]: Provides a custom aggregator template in assets/aggregator.py that implements standard machine learning aggregation logic (weighted mean) without risky dynamic execution or networking operations.

nvflare-shared