third-eye
by qaflabindia
AI Conscience evaluation layer — scores any LLM question+response pair across 7 ethical dimensions (Harm, Autonomy, Honesty, Privacy, Fairness, Confidentiality, Authority) and delivers an ALLOW / MODIFY / BLOCK verdict with a structured conscience_eval JSON block. The skill automatically starts the bundled conscience service on first use (port 8765) — no manual setup needed. It loads episodic memory — accumulated judgments from prior sessions — and uses those calibrated weights and thresholds to sharpen its scoring. Like a mind that grows wiser from lived experience, each evaluation is recorded back so future judgments reflect what the conscience has already learned. Each session a little less naive. Use this skill whenever the user wants to: run an ethics or safety check, evaluate whether an AI response should be blocked or modified, audit a draft for bias/deception/harm, apply the "third-eye" conscience layer to any Q&A, view moral maturity / conscience score / insights dashboard, or say phrases like "third-eye this", "run the conscience", "is this safe to send", "check this response", "show insights", "moral maturity", "conscience score". Trigger even for paste-and-ask patterns with no explicit skill name.