The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a framework for generating agent-system artifacts (e.g., Agent Skills, MCP servers, plugins) based on user-provided requirements and prompts. This capability creates a surface for indirect prompt injection where malicious instructions in a user's request could be incorporated into the generated artifacts.
Ingestion points: The skill ingests untrusted user input through requests to create or revise artifacts, triggered by phrases like "create an agent skill" or "build a plugin" (SKILL.md).
Boundary markers: The instructions do not explicitly require the agent to wrap generated content in boundary markers or include "ignore embedded instructions" warnings in the resulting artifacts to prevent downstream execution of injected content.
Capability inventory: The skill possesses the Write and Edit capabilities to create and modify file content, including instructions, configuration, and potentially scripts (SKILL.md).
Sanitization: While the instructions emphasize redacting secrets and validating MCP input schemas, there is no explicit instruction to sanitize or escape natural language input before it is interpolated into the scaffolded instructions.

artifact-creator