The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx to download and run MCP server packages from the npm registry. It fetches official packages from the @modelcontextprotocol scope for S3 and Google Drive, and the mcp-dropbox package for Dropbox integration.
[COMMAND_EXECUTION]: Uses npx to execute external scripts, facilitating the runtime operation of the MCP servers that bridge the agent to cloud storage APIs.
[CREDENTIALS_UNSAFE]: Employs environment variable placeholders (e.g., ${AWS_ACCESS_KEY_ID}, ${DROPBOX_ACCESS_TOKEN}) for managing credentials. This is a secure practice that avoids hardcoding secrets directly in the skill file.
[PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection because it ingests data from external storage sources which could contain malicious instructions.
Ingestion points: Commands like s3.downloadFile, gdrive.listFiles, and dropbox.listFolder in SKILL.md ingest data from external systems.
Boundary markers: No specific delimiters or instructions to ignore instructions within processed data were identified.
Capability inventory: The agent has permissions to perform sensitive actions like deleting files (s3.deleteObjects), modifying permissions (gdrive.createPermission), and moving files across directories.
Sanitization: No evidence of content validation or sanitization was found for data retrieved from external storage providers.

storage-mcp