The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements Retrieval Augmented Generation (RAG) patterns in SKILL.md and templates/rag-with-vectorize.ts, which create a surface for indirect prompt injection via external data ingestion.
Ingestion points: The ragQuery function and RAG worker template process data retrieved from Cloudflare Vectorize and user-supplied queries.
Boundary markers: The skill uses basic text labels (e.g., 'Context:', 'Question:') to delimit untrusted content within the prompt, which helps the model distinguish data from instructions but does not provide a robust security boundary.
Capability inventory: The skill is restricted to network operations via fetch to official Google APIs and Cloudflare services. It does not possess file system write access or shell execution capabilities.
Sanitization: There is no evidence of filtering or sanitizing the retrieved context for potential injection strings before it is interpolated into the final prompt sent to the LLM.
[COMMAND_EXECUTION]: The skill includes a shell script scripts/check-versions.sh which uses npm view to verify package versions. This is a standard development practice and does not pose a security risk.

google-gemini-embeddings