The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data (logs, configurations, web search results) using tools like Bash, Read, and Grep. Evidence chain: (1) Ingestion points: File reads and command outputs (e.g., node -v, node_modules, curl outputs) are analyzed; (2) Boundary markers: The skill mandates redacting sensitive data and identifying hypotheses; (3) Capability inventory: Uses standard diagnostic tools (Bash, Read, Grep, WebSearch); (4) Sanitization: Specifically instructs to redact secrets and verify only non-secret data. This attack surface is managed by safety-focused instructions.
[SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected. The skill promotes security best practices such as least privilege, secret redaction, and verification of changes.

yes-md