The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the configuration of AI Agent modules (ai-local-agent:RunLocalAIAgent) which possess an inherent attack surface for indirect prompt injection when processing untrusted input.
Ingestion points: The message and files fields described in ai-agents.md and ai-agent-full-blueprint.json allow untrusted data to enter the agent context.
Boundary markers: The skill specifically instructs developers to use restore.expect.<field>.extra.aiHelp and aiInstruction metadata to provide the LLM with constraints, which acts as a platform-supported boundary mechanism.
Capability inventory: AI agents are documented as having the capability to invoke 'tools' consisting of full Make.com scenario flows, which can perform file operations, network requests (e.g., Discord), and data storage actions.
Sanitization: The skill provides a structural framework (aiInstruction and aiHelp) intended for developers to guide the AI's decision-making process safely, though it does not mandate automated sanitization scripts.
[COMMAND_EXECUTION]: The skill mentions the existence of the code:ExecuteCode module. This is a standard platform feature for running user-defined logic and is presented in a purely informational capacity without any implementation patterns that suggest misuse.
[CREDENTIALS_UNSAFE]: The documentation actively promotes secure development practices by advising against handling raw API keys or private keys directly, recommending the use of platform-managed credential request flows to ensure sensitive data is handled securely by the end user.

make-module-configuring