Skip to content

Don't check Claude's work by hand — make it prove it

← Tips Workflow & commands

Don't check Claude's work by hand — make it prove it

The strongest verification habit in the hub — after Claude produces an output, ask it to verify its own work with real commands (grep the quotes, recompute the sums, re-run the logic) and show you the output. Commands either pass or fail; assurances always pass.

beginner

After Claude produces anything — a report, a CSV, a summary, a SQL query — your instinct is to check it by eye. Better instinct: make Claude check it with commands, and show you the output.

Verify your own report: every quote must appear as an exact substring of the source file — check each one with grep and show me the result. The counts must sum to the row count. Recompute the percentages. Fix anything that fails and re-run.

Why this works when “are you sure?” doesn’t: a command either passes or fails. grep finds the exact quote or it doesn’t. Arithmetic sums or it doesn’t. There’s no “looks fine to me” in the loop — which is exactly the failure mode of asking a to reassure you instead of prove it.

The shapes to reuse:

  • Quotesgrep -F each one against the source file.
  • Numbers → recompute independently from the raw data, compare.
  • Logic (SQL, formulas) → build a tiny test dataset with known edge cases and run it. Don’t review logic; execute it.
  • Claims → for each one, quote the exact source sentence that supports it. No quote, no claim.

One more step and it’s permanent: put the verification step in the folder’s CLAUDE.md, so Claude runs it before declaring any output done — you can’t forget what you’ve automated.

This is the habit that turns “I think the AI got it right” into “here’s the command output that shows it did” — which is the sentence your boss actually wants to hear.

Next tip →

Permission modes — what Shift+Tab toggles

Press Shift+Tab in the to cycle Claude's permission mode between default (confirm everything), auto-accept edits (just watch), and plan mode (read-only). Saves dozens of confirmations once you trust the plan.