One Claude Code Tip a Day: Turn GitHub Issues Into Debuggable Tasks
Use Claude Code to decompose a vague GitHub issue into small, verifiable tasks before asking it to edit files.
This post is part of the “One Claude Code Tip a Day” series — a daily guide to using Claude Code more effectively.
A GitHub issue is rarely a task. It is usually a complaint, a partial bug report, a product wish, and a missing test case squeezed into one box. If you hand that whole issue to Claude Code and say “implement this,” you are asking it to guess the boundary of the work before it has read the code. The session may still produce a patch, but it will be a patch against a vague story.
The better habit is to use Claude Code as a task decomposer before it becomes a coder. Today’s workflow is simple: paste the issue, point Claude at the smallest relevant files, forbid edits, and ask it to turn the issue into debuggable tasks with evidence and verification attached.
The real failure: one issue, three hidden jobs
Imagine an issue in a SaaS dashboard: “Export CSV fails for filtered orders.” A rushed Claude session might jump into the export controller, change the query, and call it done. In a real codebase, that issue can hide at least three jobs: reproducing the filter state in the UI, checking whether the API receives the same filter parameters, and verifying that the CSV writer preserves the filtered result set instead of re-querying all orders.
Those jobs require different files, different commands, and different proof. Treating them as one task makes the model optimize for motion rather than evidence.
Start with an issue decomposition prompt
Open Claude Code in the repository, then give it a constrained first pass: