The Agent Skills Directory

[SAFE]: No malicious patterns, such as prompt injection, data exfiltration, or obfuscation, were detected within the skill instructions or references.
[SAFE]: The skill includes explicit safety instructions in the 'Stop Conditions' reference, specifically mandating an immediate stop if there is a risk of exposing secrets or personal data.
[SAFE]: The framework requires agents to seek user decisions for any irreversible, destructive, or financial actions, preventing unauthorized or high-risk autonomous behavior.
[SAFE]: The use of a bounded iteration budget (defaulting to five iterations) prevents potential resource exhaustion or infinite loops during task execution.
[SAFE]: The protocol emphasizes staying within defined authority and safety boundaries and requires fresh observable evidence for every completion criterion.

loop-engineering