The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill implements functionality to fetch and process content from external URLs via blink.data.extractFromUrl and the url parameter in the upload method. This is a standard feature for knowledge base ingestion.
[DATA_EXFILTRATION]: The skill is designed to upload local files and extracted text to the vendor's storage and RAG collections (blink.storage.upload, blink.rag.upload). This data movement is the primary intended purpose of the skill for building a searchable knowledge base.
[COMMAND_EXECUTION]: The skill utilizes a vendor-specific CLI tool (blink rag) to perform searches, uploads, and collection management on the host system.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection (Category 8). Malicious instructions could be embedded in documents uploaded or fetched from URLs, which are then processed by the blink_rag_aiSearch tool using an LLM.
Ingestion points: blink.rag.upload (local files/base64), blink.data.extractFromUrl (external web content).
Boundary markers: None identified in the provided instructions to isolate search results from the system prompt.
Capability inventory: Uses google/gemini-3-flash (via blink_rag_aiSearch) to generate responses based on retrieved data.
Sanitization: No explicit sanitization or instruction filtering is documented for processed text.

blink-rag