All products

Estimate GitHub work items with AI

Compare issue size with reference work and preserve the reasoning beside it

Estimate GitHub work items with AI only after defining an observable result. The estimating-work-items skill considers effort, complexity and uncertainty before comparing an issue with smaller and larger reference stories. Each individual issue receives one, two, four or eight points. Those points express relative size, not hours or a promised delivery date.

Larger work becomes an epic with independently deliverable children. Its points are the exact sum of those children, without rounding to the issue scale. Estimates are recorded in the item body, labels and an available project field. The workflow explicitly rejects using points to compare people or teams or forcing a number to fit a predetermined plan.

What it does

  • Compare issues with reference stories
  • Separate effort, complexity and uncertainty
  • Resolve vague outcomes before sizing
  • Sum epic children without rounding
  • Update estimates when scope changes
  • Record estimate reasoning and provenance

Install and use

  1. Load the skill and reference file through the documented Claude Code plugin route. Live repository work needs GitHub CLI authentication, appropriate repository read/write access and a file/shell-capable agent.
  2. For example: “Estimate this CSV-export issue against our reference stories. Show the reasoning before changing its label, and do not convert points into hours.”
  3. Review the definition of done and the three sizing dimensions. Clarify an unobservable goal or propose a bounded investigation instead of giving uncertainty a precise-looking score.
  4. Check the body, label and project field after recording. An epic without children has no total; child proposals can be made inline without installing a decomposition companion.

FAQ

Source

Made by Britt Crawford. Original skill: estimating-work-items, published on GitHub under the MIT licence.

View on GitHub