Prediction

Forecast uncertain outcomes with base rates, reference classes, calibration loops, and explicit scorekeeping.

This skill is internally consistent and appears to do only forecasting work. Before enabling setup or allowing the agent to run its first-run actions, confirm you are okay with creating a ~/prediction/ directory and a handful of markdown files (setup uses mkdir/touch/chmod). The skill explicitly warns not to store secrets or private health/legal data — follow that guidance. If you don't want persistent local memory, decline the setup step or set integration to 'paused' / 'never_ask' in the memory template. Review the created files yourself to confirm nothing unexpected is being stored, and back them up or delete them if you stop using the skill.

Type: OpenClaw Skill Name: prediction Version: 1.0.0 The prediction skill is a well-structured framework for probabilistic forecasting that manages local state in the `~/prediction/` directory. It includes comprehensive logic for question design, calibration, and scoring (Brier scores) across several markdown files (e.g., forecast-loop.md, calibration.md). The setup.md file contains standard shell commands to initialize the environment with restrictive file permissions (chmod 600/700), and the documentation explicitly warns against storing sensitive information or credentials.