Free toolkit / No signup
A better brief.
A result you can check.
Three starting points for coding, editing and decisions. Replace the brackets with your own task, run the prompt in your existing Claude setup, and check the result yourself.
These are original instructional templates, not measured before-and-after model outputs. No performance improvement is claimed. Claude usage is separate; use access you already have.
Workflow 01
Make a coding task checkable
A change is described as done, but you cannot tell what was checked.
Vague brief
Fix this bug and make sure it works.
A more specific starting point
Task: [describe the bug and expected behavior] Reproduction: [steps or a failing test] Scope: [files or component involved] Inspect the relevant implementation before editing. Explain what evidence supports your diagnosis. Make the smallest change that addresses the reproduction. Validate the original failing case and one nearby case that should still work. At the end, list: 1. What changed and why. 2. Each check actually run, its result, and any relevant output. 3. Checks not run and the reason. 4. Remaining uncertainty. Do not label a check as passed if you did not run it. If tools are unavailable, provide the exact checks for me to run and mark them unverified.
How to judge the result
Compare the final report with the actual command output and diff. A convincing explanation is not evidence that a test ran.
A prompt cannot enforce permissions or prevent an unwanted action. Use tool permissions and repository controls for hard restrictions.
Workflow 02
Give an editor a usable voice brief
A draft is readable but sounds generic and adds claims you never supplied.
Vague brief
Make this sound more human.
A more specific starting point
Edit the draft below for [audience] who needs to [take a specific next step]. Keep the meaning and factual claims within the supplied source material. Voice reference: [paste a short sample of your own writing that you have permission to use] Prefer concrete nouns, direct verbs, and sentences I would say aloud. Preserve necessary technical terms. Remove repeated points. Do not invent statistics, testimonials, results, or personal experiences. If a claim needs evidence, flag it separately instead of strengthening it. Return the edited draft, followed by a short list of substantive changes and claims I should verify. Draft: [paste your draft]
How to judge the result
Read it aloud, compare each factual claim with your sources, and ask whether the intended reader has a clear next step.
This is an editing brief, not an AI-detector bypass. A voice sample does not establish authorship or factual accuracy.
Workflow 03
Stress-test a decision before acting
You receive a confident recommendation without knowing which assumptions matter.
Vague brief
Is this a good idea?
A more specific starting point
Decision: [state the choice] Goal: [what success means] Constraints: [budget, time, available people] Evidence available: [facts and sources] Separate known facts from assumptions. Compare my proposed option with one plausible alternative and doing nothing. For each option, describe the main benefit, cost, failure mode, and what remains unknown. Do not invent numeric estimates. Identify the assumption most likely to change the decision. Suggest a small, reversible way to test it with my existing resources. End with a conditional recommendation: choose X if Y is confirmed; otherwise reconsider Z.
How to judge the result
Look for an assumption you can actually test. Verify external claims before using the recommendation.
The model can miss alternatives or reinforce your framing. Use appropriate professional review for consequential decisions.
Download all three prompts as Markdown
For Claude Code setup, see Anthropic’s best-practice guidance. These templates are independent CLSkills Hub material and are not official Claude commands.