The Agent Skills Directory

[SAFE]: The skill utilizes the official Tencent Cloud cos-python-sdk-v5 for all cloud operations. The scripts are well-structured, using a common module for authentication and error handling.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes vector data, metadata, and bucket policies provided via CLI arguments or local files.
Ingestion points: scripts/put_vectors.py (via --vectors and --vectors-file), scripts/query_vectors.py (via --filter), and scripts/put_vector_bucket_policy.py (via --policy).
Boundary markers: Absent. While the scripts do not implement prompt-level delimitation, they are executed as standalone CLI tools.
Capability inventory: The scripts are capable of file system reads and network operations via the official SDK.
Sanitization: The scripts validate that inputs are properly formatted as JSON before processing. Given that this ingestion is the primary intended purpose of the management tool, the associated risk is considered negligible for the skill's operation.
[DATA_EXPOSURE_AND_EXFILTRATION]: Credential management follows standard practices for cloud CLI tools, utilizing environment variables (COS_VECTORS_SECRET_ID, COS_VECTORS_SECRET_KEY) or command-line arguments. No unauthorized network calls or hardcoded secrets were detected.

cos-vectors