A collection of 16 posts
Use Claude Code to turn a vague GitHub issue into a verified workplan: read the linked code, split the task into checks, expose missing assumptions, and create a first patch only after the plan survives inspection.
Read more
Use Claude Code for test-driven debugging: make it reproduce the failure first, inspect the exact failing assertion or log, apply a narrow fix, then run the same test again before broadening scope.
Read more
Use Claude Code to add small guardrail tests around the seam you are about to change: read behavior first, write characterization tests, verify failure signals, then edit with confidence.
Read more
Use Claude Code to update README docs from verified project reality: inspect commands, run checks, patch only proven instructions, and review the documentation diff like code.
Read more
A field guide to starting Claude Code in the correct repository: prove where you are, inspect git state, map the active app, and prevent confident edits in the wrong checkout.
Read more
Use /cost as a debugging checkpoint: when a Claude Code session turns into a loop, measure the spend, narrow the next prompt, inspect the diff, and buy certainty instead of motion.
Read more
Use Claude Code to ship small features safely: define the visible behavior, inspect the existing path, patch the narrowest seam, and verify with tests, logs, and git diff before expanding scope.
Read more
Use Claude Code to debug from the first real error instead of the loudest symptom: reproduce, inspect the stack trace, make a narrow fix, and verify with a second pass.
Read more
Use Claude Code for refactoring by slicing one behavior-preserving change at a time, verifying each step with tests, logs, and git diff before asking for the next move.
Read more
Use Claude Code to explain an unfamiliar codebase by demanding file-level evidence, runtime checks, and a second-pass correction instead of accepting a confident tour.
Read more