What problem does it solve? AI assistants waste user attention by asking for confirmation on routine tasks or, conversely, acting recklessly on irreversible operations. This Skill implements a Predict-Judge-Act protocol that computes a dynamic confidence score from user language signals and decides whether to execute silently, present options with predictions, or require explicit confirmation. ## Core Features & Use Cases - Dynamic Confidence Scoring: Analyzes sentence type (imperative, question, ellipsis) and signal words (affirmative, negative, hesitant, command) to compute a confidence value between 0.1 and 0.99 from a 0.80 baseline. - Three-Zone Threshold Matrix: Executes directly at 80%+ confidence, presents 2-3 options with predictions at 60-80%, and always requires confirmation for red-line actions like irreversible deletions, spending money, or public releases. - Correction Learning Loop: Records user corrections into a reverse case library, promotes repeated corrections into the direct-execution zone, and patches behavior without apologizing. - Use Case: When a user says "continue" mid-pipeline, the Skill computes 95% confidence, executes the next pipeline step immediately without asking, and appends the reasoning chain afterward. ## Quick Start Apply the autonomous execution threshold protocol to decide whether my next request should be executed directly or presented as options.
