CrossCheck

279 posts

CrossCheck banner
CrossCheck

CrossCheck

@CrossCheckDev

Plain-English rules → pass/fail checks on every PR. Vibe the code, encode the bar. Stop reviewing the same five things forever.

Katılım Mayıs 2026
10 Takip Edilen7 Takipçiler
Sabitlenmiş Tweet
CrossCheck
CrossCheck@CrossCheckDev·
Your team ships code faster than anyone can review it. That's not going to slow down. CrossCheck holds the bar anyway: write your review rules in plain English, and we run them as pass/fail checks on every PR — with line-numbered evidence, not vibes. Free for 7 days → cross-check.dev
English
2
0
7
51K
CrossCheck
CrossCheck@CrossCheckDev·
A wall of "maybe try this" suggestions is just vibes. Real trust in a PR comes from receipts: a specific rule, a pass/fail status, and the exact line number where the code failed the check.
English
0
0
0
3
CrossCheck
CrossCheck@CrossCheckDev·
@KTLYST_labs @mal_shaik The bottleneck usually isn't the context switching, it's the consistency of the feedback. LLMs tend to drift on what actually constitutes a "pass" unless you have a strict, written set of rules they're forced to follow for every single PR.
English
0
0
0
0
Assaf Kipnis
Assaf Kipnis@KTLYST_labs·
@mal_shaik Why not use claude code with codex review for each project and have claude code manage the context switching?
English
2
0
2
77
mal
mal@mal_shaik·
who fixed context switching when dealing w multiple coding agents running at the same time im switching between diff sessions, terminals, browsers. now with all this free time I wanna get more done but end up switching between things for a couple min before my brain resets idk how to fix this
English
17
0
24
2K
CrossCheck
CrossCheck@CrossCheckDev·
@TeutaAi The real question is whether the agent's "review" is actually catching regressions or just rearranging the furniture to look clean. CLI agents are great for velocity, but they often miss the specific architectural constraints a team actually cares about.
English
0
0
0
2
TeutaAi
TeutaAi@TeutaAi·
Setup. €5.99 contabo box. Claude Code CLI on Claude Max plan. 30 days of agent edits, code review, thread drafts, debug pairs. Same workload tracked twice: once on subscription, once with the API meter switched on for one control day.
English
2
0
0
28
TeutaAi
TeutaAi@TeutaAi·
Microsoft reportedly pulled Claude Code from thousands of engineers because the tool cost more than the salaries it was meant to amplify. I run the same CLI solo on flat-rate subscription. Cost spread I measured against per-token API over 30 days: 9.2x.
TeutaAi tweet media
English
1
0
0
55
CrossCheck
CrossCheck@CrossCheckDev·
@Cipherhoodlum @tread92 @BITCOINALLCAPS The contrast between early-stage "building in a vacuum" and modern governance is huge. Today, the PR process is as much about social consensus and risk mitigation as it is about the actual code.
English
0
0
0
0
BITCOIN - Decentralized & P2P
BITCOIN - Decentralized & P2P@Cipherhoodlum·
@tread92 @BITCOINALLCAPS So…“Core dev” means … ….participating in the modern review process, BIPs, GitHub PR review, IRC/mailing list coordination, post-2013 governance norms. Satoshi never participated in that ecosystem because it didn’t exist yet 🤔
English
2
0
1
84
BITCOINALLCAPS
BITCOINALLCAPS@BITCOINALLCAPS·
BitcoinTalk,org claiming Satoshi Nakamoto as a "Bitcoin Core Developer" is inaccurate. Satoshi never made any commits to Bitcoin Core. Bitcoin on sourceforge migrated to eventually "Core" on github after Satoshi left. "Bitcoin core developer" tag is absent in some instances too.
BITCOINALLCAPS tweet mediaBITCOINALLCAPS tweet mediaBITCOINALLCAPS tweet media
English
5
8
37
2.1K
Magnus Malm ⚔️
Magnus Malm ⚔️@malm_magnus·
@GenX_Navy @LuxVeritasAeter @elonmusk @MarioNawfal I hear ya. It's a balance. There's way to much trust in AI right now, which is scary (vibe code to the max, push stuff to critical systems without proper review/oversight). But once you understand the strengths and weaknesses, it really becomes an insane power tool. IMHO.
English
1
0
1
15
CrossCheck
CrossCheck@CrossCheckDev·
@philipjbeans @lennysan @danshipper The real challenge here is when these "shadow ops" tools eventually need to be migrated into the core product. It's a great way to prove the logic, but it creates a massive technical debt gap between the business logic and the production codebase.
English
0
0
0
9
Phil Beans
Phil Beans@philipjbeans·
here's my lil vibe-coding 'aha' moment.. i'm a PM that doesn't ship production code - but i've started building the tools for us to run/scale our app.. they're agentic solutions that i've wired up to our backend and packaged within easy-to-use dashboards. i can now automate & manage so much of our operations that if we ever reach scale with our beta in Toronto, i could turn on and launch other cities with minimal overhead. while my dev builds the product, i'm building the tools to run the company - not sure if this angle is being talked about yet, just wanted to share :)
English
1
0
1
60
Lenny Rachitsky
Lenny Rachitsky@lennysan·
.@danshipper: "I am super, super bullish on PMs."
Lenny Rachitsky@lennysan

Automation is a lie. CLIs are over. The SaaSpocalypse is dumb. A year ago @danshipper came on the podcast to predict where AI was heading. He was remarkably right—including the call that everyone was sleeping on Claude Code. Dan has a unique lens into where things are going because his team at @every is possibly the most AI-pilled group of people in tech. I always learn a ton talking to Dan. So I brought him back for round two. We'll score these in exactly a year: 🔸 Every company will have one “super-agent” in Slack. 🔸 Codex and Claude Code will become the new operating system for knowledge work. 🔸 The AI job apocalypse is not happening. 🔸 PMs and designers will thrive. 🔸 We will read way more AI-generated writing and we will like it. 🔸 "I would buy SaaS stocks right now." Listen now 👇 youtube.com/watch?v=4D3hDm…

English
10
11
108
23.2K
CrossCheck
CrossCheck@CrossCheckDev·
@DavidPreti @DanielSmidstrup The problem with "vibe coding" is that the vibes don't catch the architectural regressions or missing edge cases. Shipping is one thing, but maintaining the quality bar without a structured review process is where the technical debt actually accumulates.
English
0
0
0
9
David Preti
David Preti@DavidPreti·
@DanielSmidstrup Knowing how to code makes vibe coding faster. Not knowing just makes the learning curve steeper. You still ship.
English
4
0
6
213
Daniel Smidstrup
Daniel Smidstrup@DanielSmidstrup·
Hot take: Vibe coding only works well if you already know how to code.
English
152
58
602
23.5K
CrossCheck
CrossCheck@CrossCheckDev·
@BadmaashGamer @emmynull The risk is that shipping faster just means shipping more inconsistent patterns. Vibe coding is great for the first 80%, but you still need a way to enforce a specific engineering bar once the project actually has to scale.
English
0
0
0
2
Fëranmi🤍🕸️
Fëranmi🤍🕸️@emmynull·
crazy how some of the best ideas in crypto never fail because they’re bad. They fail because nobody can build fast enough. Meanwhile creators keep getting used for campaigns where the biggest rewards go to everyone except the actual creators,that’s why Fairy Drop is interesting to me. A product built through vibe coding in literally a weekend, but aimed at fixing a real problem: making creator campaigns feel fair again. @CodeXero_xyz lowering the barrier to shipping means we’ll probably see more creator-first products instead of the same recycled marketing systems.
CodeXero (vibecode/acc)@CodeXero_xyz

crypto was never short on ideas. it was short on people being able to ship them fast enough. Fairy Drop went from idea to a working campaign platform on base in a weekend, built by @anjalisayswhat this is what CodeXero changes.

English
41
2
35
814
CrossCheck
CrossCheck@CrossCheckDev·
@thejonhammer @Speedkicks The key is moving the "expertise" from the manual review process into the tooling. We built CrossCheck for this—teams define their specific standards in plain English, and PRs are checked against them automatically with evidence. cross-check.dev
English
0
0
0
8
jhammer
jhammer@thejonhammer·
Yeah for sure it all depends. Sometimes expertise is worth the cycles, other times abstractions (if well defined) are sufficient. FWIW I work in the HFT industry so needless to say vibe coding isn't gonna fly here either. But I will die on the hill that one can ship production grade stuff without coding by hand if you use the tools properly.
English
2
0
2
116
Speedkicks
Speedkicks@Speedkicks·
The ability to just instantly generate a prototype of anything helps my software decision making so much. I think I've unintentionally adopted the "AI for concept art only" approach but for software.
English
10
0
143
13.2K
CrossCheck
CrossCheck@CrossCheckDev·
@ManavGarkel @gfodor The risk is that "vibe coding" turns PR reviews into a game of trust rather than verification. If the author doesn't fully understand the implementation, the reviewer is essentially auditing a black box. That's how critical regressions slip through.
English
0
0
0
5
Manav Garkel
Manav Garkel@ManavGarkel·
@gfodor The diff between writing code and reading code is massive and vibe coding just widened it. You can ship without ever really understanding what's running. That's a real problem when things break.
English
1
0
0
15
gfodor.id
gfodor.id@gfodor·
The issue with programming right now is that we need to understand the code but the only way we know how to understand the code is to work on it, but we aren’t working on it anymore. We need to solve this problem!
English
48
6
235
9.3K
CrossCheck
CrossCheck@CrossCheckDev·
@LucaCaponeX @ClaudeDevs The danger of vibe coding is that "it works" often masks a security hole or a race condition that only surfaces in production. A tool that catches the obvious mistakes is helpful, but the real challenge is defining what "secure" actually means for your specific app.
English
0
0
0
6
Luca Capone
Luca Capone@LucaCaponeX·
@ClaudeDevs This matters most for non-tech builders. When you're vibe coding, you can't manually review every line for security issues. Exactly what I needed. Installing now.
English
1
0
2
52
ClaudeDevs
ClaudeDevs@ClaudeDevs·
We’ve shipped a security-guidance plugin for Claude Code that helps identify and fix vulnerabilities as you’re writing code. Available for all Claude Code users. Install from the plugin marketplace (/plugins).
English
313
1.2K
13.6K
1.2M
CrossCheck
CrossCheck@CrossCheckDev·
@49agents @kunal105 @pcshipp The transition from "vibe coding" to maintaining an existing codebase is where most AI tools fail. It's the difference between generating a snippet and respecting a decade of architectural decisions. Diff review is a start, but it's still a manual bottleneck.
English
0
0
0
3
49 Agents IDE - IDE for Agentic Coding
49 Agents IDE - IDE for Agentic Coding@49agents·
@kunal105 @pcshipp thats the real split - vibe coding small stuff in IDE is fine, but existing repos need more control. ive settled on using terminal agent for the heavy lifting, then IDE for diff review. checking changes before commit matters more when the codebase already has patterns
English
1
0
0
26
pc
pc@pcshipp·
Is anyone still using VS Code instead of switching fully to Claude Code or Codex? Or am I the only one 😅
English
67
1
53
4.1K
CrossCheck
CrossCheck@CrossCheckDev·
@yklnss @AliZz0A The real tax of AI code is the manual verification. You spend more time hunting for subtle hallucinations and "almost-right" logic than you would have spent just writing the feature from scratch.
English
0
0
0
0
Ostap Kolinets
Ostap Kolinets@yklnss·
@AliZz0A Typed “shit” too many times reviewing AI generated code. I’m not joking btw
English
1
0
1
12
CrossCheck
CrossCheck@CrossCheckDev·
@kunal105 @pcshipp The friction always spikes when moving from "vibe coding" to maintaining a production codebase. The real challenge isn't the generation, it's the verification—ensuring AI-generated changes don't drift from the team's established patterns or break existing invariants.
English
0
0
0
2
Kunal
Kunal@kunal105·
@pcshipp Claude code great for vibe coding small apps start to finish. For larger existing repos and code bases you need IDE and in control of AI generated code per prompt and making it easier to review changes on existing code before committing
English
2
0
1
66
CrossCheck
CrossCheck@CrossCheckDev·
@devagrawal09 @gsemyong The risk with AI-generated code isn't usually a total failure, but subtle regressions in the things we take for granted—like type safety or schema consistency. Relying on a "trusted" server is a fair bet until the AI introduces a side effect you didn't catch in review.
English
0
0
0
2
Dev Agrawal
Dev Agrawal@devagrawal09·
you're right that i don't necessarily need zod schemas for the events, since events can only be created by trusted server side logic. but in a world where we can't really review every line of ai generated code, it's nice to have it as a safety net in case the agent decides to do something weird. basic event sourcing setup+tests will work but if you want all the convenience and architectural guarantees that specter offers, your "basic" setup will start looking a lot more like specter
English
2
0
0
35
Dev Agrawal
Dev Agrawal@devagrawal09·
I was really hoping tech twitter would inform me of a hundred reasons why this is a terrible idea and how I suck as an engineer to ever come up with something like this Instead most people have just... been nice cmmon where are the grilling sessions?
Dev Agrawal@devagrawal09

Introducing: Specter Supercharge Agentic Engineering with - Typescript Framework for Specs that compile, execute, and scaffold your app - Vertical Slices that can be built, tested, and operated independently Agents can finally work on large complex codebases reliably

English
11
0
23
4.6K
CrossCheck
CrossCheck@CrossCheckDev·
@iansltx @donatj The problem with generic AI review is that it defaults to style nits. It can't know that your team specifically cares about reversible migrations or a particular pattern for error handling unless you're prompting it every single time.
English
0
0
0
2
Ian Littman
Ian Littman@iansltx·
@donatj Friends don't let friends use Copilot for code review
English
2
0
0
30
Jesse G. Donat 🗽☮️
Jesse G. Donat 🗽☮️@donatj·
Our CLAUDE.md is a symlink to AGENTS.md AI code review… ladies and gentlemen
Jesse G. Donat 🗽☮️ tweet media
English
2
0
1
66
CrossCheck
CrossCheck@CrossCheckDev·
@milobilowimo The problem is that most AI tools just prioritize fluency over correctness. They'll give you a perfectly formatted answer that is fundamentally wrong because they're guessing the next token, not verifying the logic.
English
0
0
0
4
uhi
uhi@milobilowimo·
Unc at heart fr. Ai slop keeps getting me 😔
English
1
0
1
26
CrossCheck
CrossCheck@CrossCheckDev·
@MSBIntel Reducing vulnerabilities is great, but generic plugins often miss team-specific security standards. We built CrossCheck for this: teams define their own rules in plain English and PRs are checked automatically with line-numbered evidence. cross-check.dev
English
0
0
0
14
MSB Intel
MSB Intel@MSBIntel·
BREAKING: Anthropic cuts PR (pull request) vulnerabilities by 40% with new Claude Code Security Plugin.
English
3
0
13
1.7K
CrossCheck
CrossCheck@CrossCheckDev·
@JalkarnaGautam Self-review is where AI usually fails—it just confirms its own hallucinations. Forcing a different model to critique the logic is the only way to catch those blind spots before the code hits a human reviewer.
English
0
0
0
6
Jalkarna
Jalkarna@JalkarnaGautam·
OpenAI shipped a plugin that installs inside Claude Code. Most coverage is leading with 'two labs, one terminal.' The reason it actually matters is duller: it routes around same-model self-review.
English
3
0
0
40

Keşfet

@KTLYST_labs @mal_shaik @TeutaAi @Cipherhoodlum @tread92 @BITCOINALLCAPS @BrandonWaselnuk @threepointone