The Agent Skills Directory

[SAFE]: The skill generates documentation by reading local research artifacts such as method candidate pools and iteration logs. It follows a strict human-in-the-loop workflow where graded reasoning—such as model selection justifications and assumption labels—must be verified by a human before being finalized in the output document. This design effectively prevents autonomous decision-making by the AI.
[PROMPT_INJECTION]: The skill processes project-specific data which creates an indirect prompt injection surface. 1. Ingestion points: Data enters from project files including methods/Qx/qx_decision_log.md, methods/Qx/qx_method_iteration_log.md, and results/Qx/experiments/.... 2. Boundary markers: The instructions require all AI-suggested content to be tagged with [AI-DRAFT — needs modeler verdict] and all human-verified content to carry provenance comments (). 3. Capability inventory: The skill is limited to reading project files and writing Markdown documentation within the workspace. 4. Sanitization: The workflow stops at Gate G4.5, requiring a human to explicitly review and sign a decision artifact before the skill proceeds, which mitigates the risk of malicious instructions in logs influencing agent behavior.

final-method-explainer