The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could potentially be manipulated to influence agent behavior.
Ingestion points: The agent reads tool names, descriptions, and execution results from external plugins via superset mcp tools and superset mcp call-tool.
Boundary markers: No specific delimiters or safety instructions are defined to help the agent distinguish between tool metadata and instructions.
Capability inventory: The skill utilizes Bash(superset mcp:*) and Bash(superset plugins:*) to perform actions in connected integrations.
Sanitization: There is no evidence of sanitization or structural validation performed on the text returned by external plugin servers.
[COMMAND_EXECUTION]: The skill uses the superset CLI to interact with external integrations. It includes explicit instructions to protect user data by using standard input for credentials, which prevents sensitive information from appearing in shell history or process management tools like ps.
[PROMPT_INJECTION]: Static analysis flagged a concealment pattern regarding the use of stdin for secrets. Manual review confirms this is a standard security best practice to prevent credential leakage in logs and not a malicious attempt to hide agent activities from the user.

integrations