SUN, OCTOBER 04, 2026
Independent · In‑Depth · Practitioner‑Tested
Claude Productivity

Six Prompts for AI-Assisted Work That Survives Review

System76's COSMIC desktop banned LLM-generated pull requests on 2 October - source code, comments and descriptions - and the stated reason was review workload rather than code quality. arXiv capped submissions at two a month the day before. The rules are now changing project by project, and the checklist COSMIC published asks for something tools cannot supply: that you understand your change and can answer questions about it. These six prompts get AI-assisted work to that standard, and tell you when to disclose.

⌨️ 6 prompts 🕐 Updated Oct 4, 2026
💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.
1
Find the Policy Before You Write Anything
COSMIC's ban arrived with no notice period. Check per project, per contribution, because this changes faster than contributor guides get read.
I am about to contribute to this project: [name, repo URL, or paste the CONTRIBUTING file].<br/><br/>Tell me what its rules are on AI-assisted contributions - whether generated content is banned, allowed with disclosure, or unaddressed. Quote the exact wording if it exists.<br/><br/>If the policy is silent, say so plainly rather than assuming permission, and tell me what the project's norms look like from its recent merged pull requests and issue discussions.<br/><br/>Also tell me what else the contribution checklist requires beyond the AI question.
2
Make Me Explain It Back
This is the whole test. If you cannot explain the diff, the policy is not your obstacle - the review is.
Here is a change I am about to submit: [paste the diff].<br/><br/>Ask me to explain it, line by line, as a reviewer would. For each part, ask why it is there, what it does, and what would break without it.<br/><br/>Do not accept a vague answer. If I say "it handles the edge case", ask which edge case and what happens on it.<br/><br/>At the end, list the parts I could not explain properly. Those are the parts I should either understand or remove before I submit anything.
3
Rehearse the Reviewer's Questions
Reviewers are rationing attention. Anticipating the first three questions is most of what gets a submission read seriously.
I am submitting this to [project]: [paste the change and its context].<br/><br/>Play a tired maintainer who has seen forty low-quality submissions this month. Ask the hardest questions you would ask, in the order you would ask them - why this approach, why not the simpler one, what did you test, does this match the existing patterns in the codebase, what did you break.<br/><br/>Be short and a little impatient, the way real review comments are.<br/><br/>After each, tell me what a weak answer sounds like and what a sufficient one sounds like.
4
Check It Against the Codebase, Not Against Best Practice
Generated code converges on generic convention. The thing that marks a patch as unread is that it looks nothing like the code around it.
Here is my change: [paste]. Here are the surrounding files and a few recent merged changes from the same project: [paste].<br/><br/>Tell me where my change does not match how this project actually does things - naming, error handling, test style, comment conventions, file layout.<br/><br/>I do not want general best practice. I want the specific house style of this codebase, including the parts of it you think are odd, because matching the odd parts is how a change gets merged.
5
Work Out What to Disclose
The last instruction matters. A disclosure engineered to pass is worse than not submitting.
For this contribution: [describe what you did and what tools you used], and this project policy: [paste it, or say it is silent].<br/><br/>Tell me what I should disclose and how. Walk through: was anything generated, was anything only edited or refactored with help, did I use a tool to understand the codebase rather than to write it, and which of those the policy actually covers.<br/><br/>Draft the disclosure line if one is needed. Keep it factual and short - no apology, no justification.<br/><br/>If the honest answer is that this contribution does not meet the policy, say that instead of helping me word around it.
6
Write the Policy, If You Are the One Receiving Them
COSMIC chose a ban and cited review workload. Whether that fits you depends on your volume, not on your opinion of AI.
I maintain [project]: [describe it, its size, how many contributors, and what the submission load looks like now].<br/><br/>Help me write an AI contribution policy. Cover: what is banned versus allowed with disclosure, whether it applies to code only or also comments and descriptions, what you ask contributors to confirm, and what the exceptions are.<br/><br/>Give me the blanket-ban version and the disclosure-required version, and tell me honestly which fits my project's actual load - a ban is cheap to enforce and costs goodwill, disclosure is the reverse.