The latest articles on DEV Community by Luccas Beck (@luccas_beck_0f6eb3c4aef34). https://dev.to/luccas_beck_0f6eb3c4aef34 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4216893%2F0467d92f-63d1-4ce4-b972-5406fd7245eb.jpg https://dev.to/luccas_beck_0f6eb3c4aef34 en Luccas Beck Fri, 30 Oct 2026 09:33:34 +0000 https://dev.to/luccas_beck_0f6eb3c4aef34/the-pre-merge-checklist-i-run-on-every-line-my-ai-agent-writes-2hp4 https://dev.to/luccas_beck_0f6eb3c4aef34/the-pre-merge-checklist-i-run-on-every-line-my-ai-agent-writes-2hp4 <p>AI coding agents are fast, and confidently wrong about 1 in 5 times. They ignore your conventions, "fix" a failing test by deleting it, and quietly refactor things you never asked about.</p> <p>The fix is not a better model. It is a system. Here is the checklist I run before any agent-written code hits <code>main</code>, plus the red flags I watch for in the agent's output.</p> <h2> Pre-merge checklist (for agent-written code) </h2> <ul> <li>[ ] I read the whole diff, not just the summary</li> <li>[ ] No files changed that I did not expect (<code>git diff --stat</code>)</li> <li>[ ] Tests cover the behaviour, and would fail if the code were wrong</li> <li>[ ] No test deleted, skipped, or loosened</li> <li>[ ] Lint, types, build pass (output seen, not claimed)</li> <li>[ ] No new dependency, or it is justified</li> <li>[ ] Errors handled; no swallowed exceptions; no silent fallbacks</li> <li>[ ] Input validated at the boundary; no secrets or PII in code, logs or fixtures</li> <li>[ ] No dead code, debug prints, commented-out blocks, stray TODOs</li> <li>[ ] Migrations and config changes reviewed by a human</li> </ul> <h2> Red flags in agent output </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Red flag</th> <th>What it usually means</th> <th>Do this</th> </tr> </thead> <tbody> <tr> <td>"This should work"</td> <td>It did not run it</td> <td>Ask for the command output</td> </tr> <tr> <td>Test changed in the same diff as the code, to pass</td> <td>Behaviour changed or bug hidden</td> <td>Revert test, re-check</td> </tr> <tr> <td>Huge diff for a small ask</td> <td>Scope creep or reformatting</td> <td>Re-prompt, tighter scope</td> </tr> <tr> <td>New helper that duplicates an existing one</td> <td>Did not search</td> <td>"Search for existing utilities first"</td> </tr> <tr> <td>Catch-all try/except</td> <td>Masks failures</td> <td>Require specific errors</td> </tr> <tr> <td>Confident claim about a file it did not open</td> <td>Hallucination</td> <td>"Cite file:line"</td> </tr> <tr> <td>Same failing fix repeated</td> <td>Stuck loop</td> <td>Stop and reset context</td> </tr> </tbody> </table></div> <h2> The single highest-value habit </h2> <p>Keep a short rules file (<code>CLAUDE.md</code> / <code>AGENTS.md</code> / <code>.cursorrules</code>) and add a "Known traps" line every time the agent makes the same mistake twice. That one section compounds more than any prompt trick.</p> <p>I put the full universal rules file plus this checklist in a free gist here:<br> <a href="https://gist.github.com/luccasbeckreistakehometest/b0095f5ef7ba7935bd5c305a4f7075b6" rel="noopener noreferrer">https://gist.github.com/luccasbeckreistakehometest/b0095f5ef7ba7935bd5c305a4f7075b6</a></p> <p>And I packaged the whole system I use, a drop-in rules file with 6 stack blocks, 42 prompts (understand to ship), 5 workflows and 4 checklists, into a kit if it saves you time:<br> <a href="https://luccasbreis.gumroad.com/l/bvhxerp" rel="noopener noreferrer">https://luccasbreis.gumroad.com/l/bvhxerp</a></p> <p>What is on your pre-merge checklist for agent code? Curious what I am missing.</p> ai webdev programming productivity