The Agent Skills Directory

[SAFE]: The skill defines a quality control workflow that requires the agent to inspect pixels and evaluate generated assets against five specific lenses: realization of idea, intent fit, craft quality, aesthetics, and reference consistency.
[SAFE]: The remediation steps involve refining prompts and updating parameters within a controlled generation environment. There are no patterns suggesting data exfiltration, unauthorized file access, or remote code execution.
[SAFE]: The instruction to 'adversarially review' is used in the context of critical quality assessment rather than bypassing safety guardrails or injecting malicious instructions.
[SAFE]: The retry loop is explicitly bounded to a maximum of two attempts, which prevents potential resource exhaustion or infinite loops.

flowy-verify