... spent, so the ceiling I thought I had wasn't the ceiling. Nothing failed, nothing alerted, and the number in the config was still correct the whole time. The part I think is getting worse rather than better: the diff was fine. Reading it, there was nothing to catch. The failure lived in a state nobody wrote a test for, and review is good at bad code and bad at missing cases. Shipping got fast, the sitting and thinking about ... ... Tikhaev(@konstantin_tikhaev): @asadmalik901  Scoping mutation testing strictly to guardrail logic makes it cheap enough to run in CI. You only need that paranoia where the failure cost is catastrophic. Comment from Konstantin Tikhaev(@konstantin_tikhaev): @asadmalik901  Diffs

... stuck. Thanks for giving rhun a try! Comment from Theodor N. Eng y(@theodor_n_engoy): I haven t tried rhun yet. Building Turnfeed with Codex, avatar cropping and likes that preserve scroll position needed browser checks as well as diff review. I d find it useful to keep the intended visible behavior and the observed check result beside a turn s changes. Could the Lifecycle panel accommodate that for example, a screenshot and a note that scroll position was actually ... ... along with others additions). Still thinking about the UX. Thanks for the input, I'll keep you posted about the progress. Comment from Vlad Shvedov(@r13): @matheus_paranhos1  heya Matheus. Thanks for trying rhun! It looks like the diff

... outside it. A selector inside a permitted file sits below every layer's resolution. Whether the blue landed on one button or on every primary button is decided by the sentence you typed and by whoever reads the diff. When your agent overshoots, where does the boundary live in your setup today: the prompt, a rule file, a path rule, a sandbox, or the person reading the diff

... between one and twenty-three items we have to fix, usually four to six, and the hard ones were caught against the pasted sources. If your agent reviews its own change, what does the reviewer get to see: the diff alone, the diff

... because a user pointed out that empty string is considered to be a valid ID. I still use agents on a daily basis. Now I write about 40% of the code manually. But I check every line of the diff and write my own test for everything related to authentication and validation myself. Where is the line for you? I would like to know if people have a rule that they apply when deciding which part of the code cannot ...

... been thrown. My tests have passed since I didn t have a test for such a scenario, which was found out when I accidentally kept myself logged in for much longer than expected. Now, I m looking into every diff around any lines related to security on a line-by-line basis regardless of how little the change is supposed to be. Have others developed guidelines as to what you never let the AI touch unattended? Comment from Siarhei Hamanovich ... ... siarheihamanovich): Reviewing every security-adjacent diff

... made Agent Diff Sentinel, a small CLI that scans a git diff

... OpenCode, with local conversation resume, model and reasoning controls, /usage allowance and live activity status. This release also adds vendor Windows installer choices for supported CLIs, a direct install entry when an agent is missing, change-block diff navigation (Alt+Up/Down), and fixes long lines crossing split-diff

... workflow whose contents changed. It couldn't see the other half: a dependency that starts running code the moment you npm install. That isn't a new file and it isn't a code change, it's a lockfile diff. Branch protection and required reviews do nothing about it, because there's nothing in there a reviewer would recognise as suspicious. Now the scan reads package-lock.json on your default branch and compares the subset of dependencies that carry an install ... ... same version was republished zero times. That zero is what lets the sharp case carry real weight instead of teaching you to skip the axis brew upgrade gfazioli/tap/octoscope What does your team actually do with a lockfile diff

... service got slow and found four ways of making an HTTP request in it. A fetch wrapper, axios, and two hand rolled retry loops. Nobody added them in one sitting. Each one arrived in a different session, in a diff that was correct on its own, and I approved all four. That's the failure mode I didn't see coming with agents. They're good at the task in front of them and they have no opinion about what ... ... repo already does, so every session starts from a blank slate even when the codebase doesn't. Bad code you catch in review. A second correct implementation of something you already have reads fine in a diff

... build SlimSnap.ai mostly with AI agents these days. This week I asked for a small change to annotations and the diff came back way too big for what I asked, so I asked why. The answer, word for word: "You're right, and it's my fault I wrote a second drawing engine instead of using the one that exists. Let me check what it takes to mount the real one." We already had a drawing engine. It wrote another ... ... whatever. What worries me is that I only caught it because I know what the app contains. Someone vibecoding their first product ships both engines and never finds out. Curious how other makers handle this. Do you review every diff

... noticed a pattern with AI coding agents that bothered me more than the actual code quality. I'd ask for a small change. The change would work. Then I'd look at the diff and realize the agent had touched: 11 files for what I thought was a 2-file task. Sometimes there was a good reason. Sometimes it was just architecture drift disguised as cleanup. So I started experimenting with a simple rule: Every task gets a change budget ... ... needs more, stop and explain why before continuing. That has created a useful checkpoint. If the agent suddenly says: I also need to refactor these 7 other files I can decide whether that is actually necessary before the diff

... AI Commit Message Tagline AI-powered git commit message generator that reads your git diff Description AI Commit Message is an open-source CLI tool that automatically generates clean, conventional commit messages from your git diff

... been using VS Code with ChatGPT Codex for development, but I keep running into a frustrating workflow bottleneck: when I ask the agent to edit or fix a tiny section of a file, it often insists on rewriting the entire codebase or file from scratch. This burns context windows, introduces random side-effect bugs, and wastes a ton of time reviewing diffs

... Think about what an agent session actually risks. It edits files in a working tree. If that tree was clean when it started, the entire session's damage is one diff you can read and one reset you can throw away. Branches cost almost nothing, so every task can live on its own one, and merging stays a human decision made after reading the evidence. Underneath that sits the layer most people only meet in a crisis: the reflog. Every ... ... write. The habits this buys: commit before you hand over the keyboard, because everything uncommitted stands outside the net (an agent deleting an untracked file performs a real, unrecoverable deletion). Give each task its own branch and review by diff

... exploring is: Finish a design or implementation in Codex Desktop. Ask Claude Desktop to open a new, visible review session in the same repository. Give Claude the relevant context from the source conversation not just the code diff. Return Claude s findings to the original Codex task automatically. To be precise about the existing landscape: Claude Desktop s Code tab runs Claude Code and supports plugins. OpenAI s codex-plugin-cc already covers much of the Claude Desktop / Claude Code ... ... built an early open-source macOS implementation to test this: https://github.com/WarrenJones/d... I m curious: Do you use a second coding agent for independent review? Does the reviewer need the source conversation, or is the repository and diff

... Vibecoding gave me speed, and quietly took something back. A clear picture of what I'm actually building. I can see every diff. The AI ships feature after feature. But when I zoom out, I've lost the thread. What's really in here now, how the pieces fit, whether I'm still heading where I meant to go. The commits tell me what changed. Nothing tells me what it means, or where it's quietly drifting. When I wrote ... ... head for free. Now the code moves faster than my understanding of it, and the gap keeps widening. So for people building fast with AI: what keeps you oriented? A tool, a ritual, or do you just trust the diffs

... twice. Which it's allowed to do. Then the handler credited the customer twice. Why it slipped through The bug wasn't on any line. It was the assumption between the lines: every event arrives exactly once. Reading the diff can't catch that. My tests didn't either. I wrote the fixtures, so I never thought to send the same event twice. My assumptions were just checking my assumptions. What actually caught it Running the failure. Deliver the event ... ... third party get to do that I'm not choosing" as its own checklist item during review, separate from does the code look correct. retries, out of order delivery, partial failures. none of that shows up in a diff

... identical revert reproduces the exact OID. detection is a lookup, not a new mechanism. The real question is how far back to keep it a baseline already stops scoring past 30 days, since by then most of the diff is my own commits. worth conceding too, the scan is on-demand rather than a cron, so the sampling window is whenever someone presses the key. sharpens your point rather than softens it. anyway, really glad you wrote this one. "nothing ... ... formatting change, not a mechanism. "a fixed lookback is a published dwell time" is the line I'll be quoting back at myself, tbh. thirty days got picked because past it most of the diff

... times over three days once at dinner, once from my phone, once from bed and the whole thing showed up as a single pull request against main. The agent did the engineering. I never read a diff. I never touched my keyboard for it after the first sentence. That is not how handing a big feature to an AI usually goes. So let me show you the actual machine the blueprint I approve, and the loop that grinds under ... ... babysitting tax Hand a real feature not a one-line bug fix, an actual  feature   to an agent that can write and ship code, and you get one of two deals. Babysit every diff

... week we reinvent the company. Here s are my favourite things we shipped last week: ( quick video here ) - JSON prompts (turn any prompt into JSON format in a click) - Prompt Notes (add notes of what a prompt is about) - diffs (see what changed when editing a prompt) Should we add Skills? Pretty Skills sounds pretty good Can't wait to hear your thoughts ...

... instead of from whatever survived in context. Gal's point about a second model not knowing the first one's choices stops applying once the choices live somewhere any model can read. It does nothing for the half read diff. That part I still have to read like everyone here says. Comment from Abdullah Javaid(@abdullah_javaid3): @raffay_sajjad  That's Rabnoor's third record type showing up in practice. Not a decision, not reasoning, an unresolved argument ... ... unresolved: X" is one line and survives, which beats losing it entirely. Comment from Alexander Bickov(@bickov): @thenameisarian  Rate limit as forced checkpoint, not a delay, that's basically what I've landed on too. Read the diff

... cannot answer cleanly for our own product: If your agent gets one thing wrong today, what is the largest number of people who find out before you do? For a coding agent that is usually one. You, reading a diff. The moment the agent sends, books, charges, replies to a customer or touches production, that number stops being one, and most of us have never actually worked out what it is. So whether you run agents or ship them ... ... because it's dumb, but because the radius is determined by capabilities it isn't even looking at. That's the strongest version of the "worst judge of its own reach" point I've seen. And your authority-armed-versus

... size, but the fix is not a better denominator, it is holding the stopping condition constant in the numerator. And "done" has to be an external check rather than the model deciding it is finished. A test suite, a diff review, anything, as long as the tool does not get to grade its own work. The reason nobody publishes this is that it destroys the clean headline. The 5.5x stops being a number and becomes a distribution ...

... Mine's anything touching payments or auth. Not because the code would be wrong, because when it's wrong there's no diff to review before it already happened, a webhook fires and the consequence is already out the door. Curious where other solo builders draw that line, since "just review the diff