What problem does it solve? Estimates for discrete tasks are rarely checked against actuals, so estimation accuracy never improves. This Skill captures start and end timestamps alongside both a gut estimate and a calibration-factor-adjusted estimate, building a per-task-class dataset that reveals whether estimates and the calibration formula itself are improving over time. ## Core Features & Use Cases - Dual-estimate logging: Records both a gut estimate and a calibrated estimate (gut × class median ratio) for each task, with a minimum of 5 completed class entries required before calibration activates. - Start and end modes: /time-task start generates a UUID and logs the estimate; /time-task end computes actual minutes, gut_ratio, and calibrated_ratio. - Task-class taxonomy: Supports classes like skill-authoring, research, debug, migration, and substitution-edit so calibration factors are computed per class. - Use Case: Before starting a research task you estimate 30 minutes; the skill finds 21 completed research entries with median ratio 0.44, logs a calibrated estimate of 13 minutes, and at completion reports both ratios so you can see which estimate was closer. ## Quick Start Ask the AI to start timing a task by saying "time this task: research the API options, estimate 30 minutes, class research" and later ask it to end the task with the returned task_id.
