Home

An AI skill for estimating AI-assisted and hybrid human+agent development work.

Early development. Formulas, multipliers, and default values are being actively calibrated and may change between versions.

---

Quick Navigation

Getting Started

• Installation — Setup for Claude Code, Cursor, Copilot, Windsurf, Cline, Aider, and more
• Quick Start — Your first estimate in 2 minutes
• Modes and Paths — Quick vs Detailed, Single vs Batch
• Cooperation Modes — Human-only, Hybrid, or Agent-first

Core Concepts

• How It Works — The computation pipeline explained
• Task Types — Multipliers for different kinds of work
• Agent Effectiveness — Why larger tasks need more human effort
• Confidence Levels — Expected vs Committed estimates
• Cone of Uncertainty — How estimate ranges narrow over time
• PERT Statistics — Three-point estimation and confidence bands
• Small Council — Subagent validation for M+ tasks

Reference

• Formulas — Complete arithmetic, lookup tables, and JSON schema
• Questionnaire — All intake questions and what they feed
• Token Estimation — Token consumption and API cost estimation
• Tracker Integration — Linear, JIRA, ClickUp, GitHub Issues, Monday, GitLab, Asana, Azure DevOps, Zenhub, Shortcut
• Output Format — Summary-first ordering, batch tables, warnings

Improving Accuracy

• Calibration — Tuning estimates with actuals
• PRED 25 — The primary accuracy metric
• Reference Stories — Anchoring estimates with real examples
• Anti-Patterns — Common estimation mistakes and how to avoid them
• Sprint Velocity — Mode-aware velocity tracking
• Deep Validation — 110k data points, 11 analyses, Parameter Audit Card

For Contributors

• Evals — Testing the skill with the Claude Skills 2.0 framework
• Calculator Generation — Generating standalone scripts in any language
• Research — Papers and studies behind the formulas
• Contributing — How to help improve the skill