Daniel Salazar

SQL Server tuning expert company. SAP reporting from Excel using vba. Data collection with opendatakit (ODK) program android app. Cryptocurrencies. Ethereum.

[Daniel Salazar @dsalazar]

@benkimbuilds Yo

[Daniel Salazar @dsalazar]

@theodormarcu I'd like to try it

[Daniel Salazar @dsalazar]

@adrianmg @burkeholland I use it.

[Daniel Salazar @dsalazar]

@WorkflowWhisper PLAYBOOK

[Daniel Salazar @dsalazar]

@zenorocha I'm doing a forms creation i added resend as a way to send the results back to the user. May be nice to test it. https://t.co/umKkD72qER

[Daniel Salazar @dsalazar]

Yes, I've had luck with Claude on 100k+ LOC projects. Key tip: First pass—ask it to identify only necessary files, plan very localized changes, no refactors/unnecessary tweaks. Keep it simplest; for new stuff, create isolated objects w/ minimal existing code mods. Second pass: Implement the plan.

[Daniel Salazar @dsalazar]

Not enough is being done about the psychological & physical effects of sugars/carbs on satiety for some folks. I lost over 75kg (~165 lbs) in 5 years on keto—I'll stay on it forever or face that brutal hunger 2 hrs after carbs. It's not just mental; something physical ramps up cravings, portions balloon, & you overeat. Not pleasant at all.Calories in/out? Nah, you can't control it when that hunger hits after being heavy. GLP-1 drugs will be a game-changer for many—it's HARD to eat only what's needed. 15+ yrs on keto: zero hunger issues vs. carbs where it strikes fast & fierce. Physically, it's awful—zero to do with willpower.

[Daniel Salazar @dsalazar]

Look at what ironspeed used to do. It was amazing. I made a lot of crud apps with it. It generated full crud apps starting with the db defined by the programmer and from that built the whole app and allowed you to modify it. I started doing something like that a while ago using the llms.

[Daniel Salazar @dsalazar]

@rork Rork

[Daniel Salazar @dsalazar]

I had this issue for a while and this is how codex finally helped me solve it without reformatting that Mac.

🧵 Fixing the Stubborn macOS iCloud Authentication Loop (The Pro Guide)

If your Mac is stuck in an endless loop asking for your iCloud password, the problem is usually a broken keychain token or corrupted account cache. I finally fixed mine using steps guided by AI.
Here's a 3-Phase technical tutorial:

🔍 Phase 1: Clear Corrupted iCloud Tokens

The primary fix is deleting the old, stale authentication tokens.
- 𝗜𝗱𝗲𝗻𝘁𝗶𝗳𝘆 𝗧𝗼𝗸𝗲𝗻𝘀: Open 𝗞𝗲𝘆𝗰𝗵𝗮𝗶𝗻 𝗔𝗰𝗰𝗲𝘀𝘀 $\rightarrow$ Search for icloud. You are looking for entries with names like https://t.co/VRWZxwpedo.account.AppleIDAuthentication.token.
- 𝗦𝘁𝗼𝗽 𝗔𝗴𝗲𝗻𝘁𝘀: Open 𝗔𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗠𝗼𝗻𝗶𝘁𝗼𝗿 $\rightarrow$ Force Quit accountsd, cloudd, and apsd to unlock the tokens.
- 𝗗𝗲𝗹𝗲𝘁𝗲 𝗧𝗼𝗸𝗲𝗻𝘀 (𝗖𝗟𝗜): Run these commands in 𝗧𝗲𝗿𝗺𝗶𝗻𝗮𝗹 (enter your password when prompted):Bashsecurity delete-generic-password -l https://t.co/VRWZxwpedo.account.AppleIDAuthentication.token ~/Library/Keychains/login.keychain-db security delete-generic-password -l https://t.co/VRWZxwpedo.account.AppleIDAuthentication.token-expiry-date ~/Library/Keychains/login.keychain-db
- 𝗥𝗲𝗯𝗼𝗼𝘁 & 𝗧𝗲𝘀𝘁. If iCloud connects, you're done!

🗑️ Phase 2: Rebuild Account Cache (If Phase 1 Fails)

If the loop continues, the user account database needs a refresh.- 𝗘𝗻𝗮𝗯𝗹𝗲 𝗙𝘂𝗹𝗹 𝗗𝗶𝘀𝗸 𝗔𝗰𝗰𝗲𝘀𝘀: Go to System Settings $\rightarrow$ Privacy & Security $\rightarrow$ Full Disk Access $\rightarrow$ Add 𝗧𝗲𝗿𝗺𝗶𝗻𝗮𝗹. Relaunch Terminal.
- 𝗕𝗮𝗰𝗸𝘂𝗽/𝗠𝗼𝘃𝗲 𝗙𝗶𝗹𝗲𝘀: Run these commands to move the corrupted account caches (they will be automatically rebuilt):Bashmv ~/Library/Accounts ~/Library/Accounts-backup mv ~/Library/Application\ Support/iCloud ~/Library/Application\ Support/iCloud-backup mv ~/Library/Caches/com.apple.iCloudHelper ~/Library/Caches/com.apple.iCloudHelper-backup
- 𝗥𝗲𝗯𝗼𝗼𝘁 & 𝗧𝗲𝘀𝘁.

🔑 Phase 3: Reset Login Keychain (The Final Resort)

If all else fails, a full keychain reset is needed. ⚠️ 𝗪𝗔𝗥𝗡𝗜𝗡𝗚: 𝗧𝗵𝗶𝘀 𝗱𝗲𝗹𝗲𝘁𝗲𝘀 𝗮𝗹𝗹 𝗹𝗼𝗰𝗮𝗹 𝗮𝗽𝗽/𝘄𝗲𝗯𝘀𝗶𝘁𝗲 𝗽𝗮𝘀𝘀𝘄𝗼𝗿𝗱𝘀 𝗻𝗼𝘁 𝘀𝘆𝗻𝗰𝗲𝗱 𝘁𝗼 𝗶𝗖𝗹𝗼𝘂𝗱 𝗞𝗲𝘆𝗰𝗵𝗮𝗶𝗻. 𝗬𝗼𝘂 𝗺𝘂𝘀𝘁 𝗿𝗲-𝗲𝗻𝘁𝗲𝗿 𝘁𝗵𝗲𝗺.- 𝗥𝗲𝘀𝗲𝘁 𝗞𝗲𝘆𝗰𝗵𝗮𝗶𝗻: Open 𝗞𝗲𝘆𝗰𝗵𝗮𝗶𝗻 𝗔𝗰𝗰𝗲𝘀𝘀 $\rightarrow$ Preferences $\rightarrow$ "Reset My Default Keychain...". This creates a clean login.keychain-db and backs up the old one as login-renamed-....
- 𝗥𝗲𝗯𝗼𝗼𝘁 & 𝗦𝗶𝗴𝗻 𝗜𝗻: Restart the Mac and sign into iCloud. It should succeed now!
- 𝗥𝗲𝗰𝗼𝘃𝗲𝗿 𝗣𝗮𝘀𝘀𝘄𝗼𝗿𝗱𝘀: To get essential saved passwords back: In 𝗞𝗲𝘆𝗰𝗵𝗮𝗶𝗻 𝗔𝗰𝗰𝗲𝘀𝘀 $\rightarrow$ File $\rightarrow$ Add Keychain $\rightarrow$ Select the old login-renamed-... file. 𝗢𝗻𝗹𝘆 𝗱𝗿𝗮𝗴 𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗽𝗮𝘀𝘀𝘄𝗼𝗿𝗱𝘀 you need (like Wi-Fi, local apps) from the old keychain into your new "login" keychain.

[Daniel Salazar @dsalazar]

An old Mac Pro has been locked out of iCloud for ages. Didn't want to format, and all my Google/Claude/etc. searches failed... until tonight! 🔥 Codex walked me through deleting some keychain entries. It worked!

The Lesson: If an AI can't solve your problem today, try again in a few months. They are accelerating at warp speed! 🤯

[Daniel Salazar @dsalazar]

Excited to share a little project I whipped up a few months ago using Python and some vibecoding vibes: MiLumoHomeKit!

This nifty HomeKit bridge seamlessly integrates your Philips Hue lights and devices from voice assistants like Alexa and Google Home into Apple's HomeKit ecosystem. Now, you can control everything—from legacy Hue bulbs to non-native smart gadgets—right from Siri, the Home app, or your iPhone, Apple Watch, and Mac. It's all powered by Python and runs smoothly on a low-power setup like a Raspberry Pi (even the Zero 2 W), keeping your smart home centralized and always on.

Why did I build it? Mainly as a vibecoding experiment to breathe new life into old devices. When Philips discontinued support for the original Hue Bridge (V1), it lost Siri integration and modern platform compatibility. MiLumoHomeKit fixes that, emulating a HomeKit device to revive those forgotten bulbs and hubs. Plus, it pulls in non-HomeKit gear via Alexa/Google, giving you a unified Apple-centric control hub.

The best part? It's been running flawlessly at home for months now—no hiccups! I use it daily to manage all my revived devices, and it's made my smart home feel fresh again. If you're dealing with outdated Hue stuff or want to unify your ecosystem, give it a spin!

Check it out on GitHub: https://t.co/NNkmOQPeIU

What cool hacks have you done to resurrect old tech?

[Daniel Salazar @dsalazar]

Stop hunting for that sneaky extra closing bracket } or parenthesis ) an LLM added! It's a 🔒 loop for them and a time sink for you.

📷Actionable fix: Pass the whole code file to Gemini Flash or Gemini Pro. Context is key—they'll resolve file-level syntax errors fast!

[Daniel Salazar @dsalazar]

This collab boosted our efficiency, but it reminded us: LLMs need your expertise to thrive on tough tasks. What's your LLM debugging story?🛠️ Debugging Complex Key Events with LLM Assistants: A Critical LessonWe recently tackled a pesky Enter and Shift-Enter bug in our chat client's Quick Action HUD. The process highlighted a key flaw in over-relying on LLMs for intricate event-chain debugging—without human context, they can lead you astray.The Bug & The AI PitfallThe problem: With the suggestion HUD visible, Enter (to send typed messages) and Shift-Enter (to send selected suggestions) failed inconsistently. Earlier fixes for a focus-loss issue (which risked sending partial messages) had suppressed the Enter key https://t.co/iiQS313tii Stumbles: Tools like Codex and early Gemini Pro got confused on event handling. Codex suggested two fixes in the wrong code section, misunderstanding where keypresses were processed.
Human Breakthrough: The game-changer? A classic debugging session. The developer set Xcode breakpoints in the AI-suggested functions—and they never triggered. This revealed the issue wasn't there, but in a higher-level macOS Responder Chain event handler.
Armed with this insight, we refocused. Gemini Pro, now given proper context, nailed the fix by overriding keypresses at the right level.The Technical FixDitch patching low-level text edits; intercept keys at the view's event layer (e.g., keyDown: or shouldHandleKeyEvent:):Plain Enter/Return: Mark the event as "allowed," handle it to send the typed message, and dismiss the HUD cleanly.
Shift-Enter/Shift-Return: Intercept to trigger the HUD's selection logic, sending the quick suggestion.
Actionable Tips for LLM-Assisted DebuggingLLMs shine in code generation but falter on context in big codebases. Keep the human in the loop as the ultimate guide:Prioritize Breakpoints Over Blind Tweaks: If an LLM fix flops, don't iterate endlessly. Set breakpoints to verify if the code path even runs. No hit? The LLM's targeting the wrong spot—pivot immediately.
Master the Responder Chain: For key event bugs, think hierarchy. Keys can get "eaten" upstream. LLMs often zoom in on text fields; check if the issue's in the window's event flow instead.
Supply Rich Context to LLMs: Once you pinpoint the right function (e.g., via debugging), feed the LLM that specific code snippet plus your findings (like "Breakpoints skip the old function; events hit here"). This transforms it from guesser to precise collaborator.

[Daniel Salazar @dsalazar]

Speed Wins: Fast AI Feedback Loops Accelerate Development.

In modern software development, raw coding power matters less than how quickly you can iterate. Based on real-world usage, two AI tools stand out for speed:

Gemini Pro excels at rapid bug diagnosis in large codebases—if you send the full context at once. Using the CLI, explicitly instruct it to “read the entire file/repository” to avoid slow, piece-by-piece processing.

Grok (via OpenCode) is exceptionally fast at validating that your implementation matches your plan and catching regressions. While not as strong as Claude or Codex for writing complex code, it’s a top-tier reviewer for alignment and correctness.

This aligns with the OODA Loop, a key insight from U.S. military strategist Col. John Boyd: it’s not the quality of a single decision cycle (Observe, Orient, Decide, Act) that drives progress, but the speed of repeated cycles. By cycling faster than opponents, you gain an edge—much like rapid AI feedback outpaces slower methods.

In practice: use fast reviewers like Gemini and Grok to shorten your loop, and reserve heavy coders like Claude and Codex for when you need depth. Speed isn’t just convenience—it’s your main lever for improvement.

[Daniel Salazar @dsalazar]

Woohoo, new record. 39 minutes of uninterrupted Codex productivity! 🎉 For just $20 a month, this is amazing! Time to put it to the test. First, a quick pass with Grok using opencode to ensure all requirements are met, then the real fun begins: testing the new functionality. Let's see how long this takes!

[Daniel Salazar @dsalazar]

This use of GPT-5 blows past mere search—it's a true navigator, unlocking all surrounding knowledge and revealing hidden connections. Pure acceleration!

[Daniel Salazar @dsalazar]

Today, I optimized memory usage in my Node.js open-source app (backend + frontend) from a whopping 9GB RAM down to <400MB—in just one day! Here's my quick discovery guide:
- Lazy Loading: Delay component loads until needed—huge win for frontend and backend.
- Backend Cleanup: Stop loading frontend assets server-side; separate concerns!
- Other Tweaks.
AI helpers
@AnthropicAI 's Claude, Qwen, & OpenAI's Codex made it a breeze. Try it!

[Daniel Salazar @dsalazar]

@jackfriks @postbridge_ I think it was the vibecoding aspect of it. Gave you lots of reach.

[Daniel Salazar @dsalazar]

Remember when installing open-source projects was a whole weekend project?
Now I just ask Claude: "Hey, install this repo, find all its dependencies, fix what's broken, and make me a run script."
What used to be hours is now a single prompt. 🚀