The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (SAFE): The skill references the sentence-transformers and scikit-learn libraries. These are standard, trustworthy packages for NLP and machine learning tasks. No automated download or installation of untrusted scripts was detected.
[DATA_EXFILTRATION] (SAFE): The workflow focuses on local data parsing and similarity calculations. No network requests to non-whitelisted domains or access to sensitive system paths (e.g., ~/.ssh, ~/.aws) were identified.
[COMMAND_EXECUTION] (SAFE): The provided Python snippets are restricted to loading models and performing mathematical operations. There is no evidence of arbitrary command execution or use of dangerous functions like eval() or subprocess.run() on untrusted input.
[PROMPT_INJECTION] (SAFE): The instructions are purely task-oriented and do not contain phrases attempting to override the agent's safety protocols or core instructions.
[INDIRECT_PROMPT_INJECTION] (LOW): While the skill involves parsing external document data (Ingestion point: Step 1), its primary function is computing numerical embeddings for retrieval. The instructions emphasize strict parsing and validation, which reduces the surface area for processing malicious instructions embedded in data.

mteb-retrieve