The Agent Skills Directory

[PROMPT_INJECTION]: Manual review of instructions flagged for 'silently scanning' and 'concealment' confirms these are safety-oriented design patterns. They instruct the agent to internally triage for red flags like abuse, coercion, or self-harm before providing coaching, ensuring ethical responses rather than attempting to bypass safety filters.
[COMMAND_EXECUTION]: The package includes developer scripts (validate_skill.py and smoke_test.py) that use the subprocess module to run other internal scripts for verification. These calls are implemented securely using argument lists (avoiding shell=True) and only operate on local files within the skill directory.
[DATA_EXFILTRATION]: No network operations or data transmission patterns were identified. All script functions operate locally, and there is no evidence of credential harvesting or sensitive file access.
[SAFE]: The skill follows security and safety best practices, particularly in its semantic triage of sexual language, which distinguishes between consensual adult activities and potential safety risks through contextual analysis.

relationship-science-coach