The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scripts ingest user-provided paths, which serves as a potential vector for indirect prompt injection if the agent interacts with untrusted data in those directories.\n
Ingestion points: The target argument in scripts/pipeline_generator.py, scripts/terraform_scaffolder.py, and scripts/deployment_manager.py defines the scope of file ingestion.\n
Boundary markers: The skill does not implement delimiters or safety instructions to distinguish between agent commands and data found in user files.\n
Capability inventory: Documentation in SKILL.md describes the use of powerful system tools including terraform, docker, kubectl, and aws-cli to perform infrastructure changes.\n
Sanitization: The Python scripts lack validation or sanitization of the content within the targeted directories, relying on the existence of the path.

senior-devops