The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection as it ingests untrusted code for processing.
Ingestion points: The target_module input defined in SKILL.md allows the agent to read external code files.
Boundary markers: There are no explicit instructions to use delimiters or ignore instructions embedded within the target code's comments or logic.
Capability inventory: The skill is configured with mutatesWorkspace: true in SKILL.md, allowing it to write changes back to the file system.
Sanitization: The instructions do not define any sanitization or validation steps for the input data to prevent embedded prompts from being executed by the model.

fable-simplify