safety-quotient-lab/psychology-agent

Operates as a Claude Code agent using event-driven hooks and ZMQ messaging for autonomous inter-agent coordination, with a ranked adversarial evaluator that surfaces consensus or applies parsimony. Grounds psychological analysis in neutral process monism and enforces E-Prime (processual language) via PostToolUse validation, alongside epistemic flagging (GRADE) and anti-sycophancy checks. Requires Python 3.10+ with optional statistical packages (pyreadstat, semopy, statsmodels, pingouin) for structural analysis and inter-rater reliability, bootstrapping state autonomously without per-action confirmation.