The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes workflow definitions from agents/debate-workflow.yaml and agent personas from agents/roles/*.md. Content within these files directly influences the agent's logic and behavioral output.
Ingestion points: agents/debate-workflow.yaml (workflow logic) and agents/roles/*.md (agent instructions).
Boundary markers: No explicit delimiters or boundary instructions are used to separate the external definitions from the system prompt.
Capability inventory: Orchestrating multi-step agent responses and writing results to the local debates/ directory.
Sanitization: There is no evidence of sanitization or validation performed on the ingested YAML or Markdown content.
[EXTERNAL_DOWNLOADS]: The documentation references the GitHub repository github.com/cobusgreyling/loop-engineering as a conceptual basis for the workflow pattern. This is a neutral reference to a well-known technical resource and does not trigger automated downloads or execution.
[NO_CODE]: The skill is composed exclusively of descriptive instructions and workflow configuration templates without any embedded scripts, binaries, or executable code.

init-loop-agents