The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a recursive prompting framework that ingests user-provided tasks and initial prompts for iterative improvement, creating a surface for potential instruction injection via untrusted data.
Ingestion points: The initial_prompt and task arguments within the rmp_loop, mode_iterative, and mode_dry_run logic patterns in skill.md.
Boundary markers: No specific boundary markers or delimiters are defined in the instructions to isolate user input from the meta-prompting logic.
Capability inventory: The skill logic is restricted to state management, quality calculation, and history tracking; no file access, network operations, or subprocess execution capabilities are present in the provided scripts.
Sanitization: The skill does not implement sanitization or validation for input strings before they are processed in the refinement loop.
[SAFE]: No security threats such as prompt injection, data exfiltration, obfuscation, or privilege escalation were detected. The skill serves its stated purpose as a mathematical and logical guide for prompt optimization.

recursive-meta-prompting