SelimA

DevOps, Fintech, LoB Banking, Infrastructure, Enterprise Design, Observability, on-chain, OSS: @hashspandev, Benchmark enthusiast

[SelimA @Sel1mC]

Built, optimized and running:

Time per refresh: ~62 ms → ~35 ms
CPU per refresh: ~23 ms → ~13 ms
Processes per refresh: 7 → 2
20 sessions refreshing at once: ~0.46 s → ~0.26 s CPU

[ClaudeDevs @ClaudeDevs]

You can now mod Claude Code:

- Change how it behaves
- Customize the UI
- Swap in your own features

Write one with a few lines of TypeScript, or have Claude build it for you. Mods ship inside plugins, so you install them with /plugin in the CLI or desktop app.

A few examples:

[SelimA @Sel1mC]

how it was measured, plus more benchmarks: https://t.co/sceWv6LmcB

[SelimA @Sel1mC]

how many bytes does one llm call add to your traces?

same 32 KB prompt, openai python sdk, default settings:
openinference 65 KB, it stores the prompt twice
openllmetry 33 KB
langfuse 14 KB, gzip is on by default

two env vars bring openinference down to 14 KB

[SelimA @Sel1mC]

how to use it, plus more claude code add-ons: https://t.co/dl179a3lyu

[SelimA @Sel1mC]

rewrote my claude code status line from bash + jq + awk into one awk process

per refresh: 19.8 ms to 6.3 ms, cpu 13.2 to 3.5 ms, 3 processes to 1

and added two things from the statusline docs I wasn't using: a timer for when the prompt cache goes cold, and when a low rate limit resets

[SelimA @Sel1mC]

repo: https://t.co/U8QU8IJCZ5

3 read-heavy sessions on zustand, 3 runs each. it kept every message word for word and dropped every tool output, so here it acted more like "drop tool results" than smart picking. small n, not a general claim

[SelimA @Sel1mC]

tried fast-jev-compaction (7.7k★) in claude code, against the built-in /compact on the same sessions

recall after compact: 45/57 vs 46/57
time to compact: under a second vs ~23s

one catch: after --resume, the context came back bigger than before compacting

[SelimA @Sel1mC]

@jha_ruchir25475 That’s a good approach! Intelligent-tiering can make long-term log storage more cost-effective. I think it fits nicely alongside having different retention policies for different log types, especially when compliance and audit requirements come into play

[SelimA @Sel1mC]

Can a wrong "no results" turn into "it never happened"?

Now that we work alongside AI agents in our logs, it can happen more easily than we think. Let's take a quick look at why, and how we can prevent it 🙂

Log Notes #2 https://t.co/DB7dLCrk3w

[GitHub @github]

Git activity more than doubled between September 2025 and August 2026, reaching 473.3B events per month.

We’re rebuilding our Git infrastructure while GitHub keeps running, creating a foundation for agent-scale software development. https://t.co/H5J2qxCqMT

[Uber Engineering @UberEng]

https://t.co/p4OzJicyw8

[Uncle Bob Martin @unclebobmartin]

People seem shocked that the author of Clean Code doesn't read code anymore. But that should not be surprising.

Why did I write Clean Code? What is the goal of keeping code clean? To get it out of the way of the real job -- thinking.

[Can Vardar @icanvardar]

i still call it twitter btw

[SelimA @Sel1mC]

We log everything. Every request, every retry, every health check. Some months that's billions of log lines

Then one day you need one event from last March and all that just in case data is suddenly between you and the answer

What actually helps?
• drop patterns nobody queries → up to 50% less log volume
• store smarter, compression and log-specific index modes → up to 65% less storage
• fewer GB = faster answers, less for every query, search job or restore to read

A question I like asking: who actually reads this log? if the honest answer is nobody, just in case… it's probably noise

And if an AI agent is going to dig into your next incident, it's digging through the same noise. less noise, better answers right?

full article 👇
https://t.co/zuxCphtpsS

[SelimA @Sel1mC]

In regulated finance, how long we keep logs is mostly decided for us: PCI DSS, card dispute windows, DORA.

The harder question is what happens when someone asks for one specific line out of a year of them. https://t.co/pm5yuaDlrJ

[SelimA @Sel1mC]

The Tokyo Stock Exchange had to switch over within 30 seconds. On October 1, 2020, it stayed closed all day.

A short note on RTO, RPO, and why the number on paper is the easy part. https://t.co/jhFwBQ57RW

[SelimA @Sel1mC]

A year of logs doesn't help much if you can't find the one event that matters

[SelimA @Sel1mC]

In regulated finance, how long we keep logs is mostly decided for us: PCI DSS, card dispute windows, DORA.

The harder question is what happens when someone asks for one specific line out of a year of them. https://t.co/pm5yuaDlrJ

[SelimA @Sel1mC]

In regulated finance, how long we keep logs is mostly decided for us: PCI DSS, card dispute windows, DORA.

The harder question is what happens when someone asks for one specific line out of a year of them. https://t.co/pm5yuaDlrJ

[ClaudeDevs @ClaudeDevs]

Perf win of the day: Claude Code CLI now uses 2x less CPU at p99.

Bun's garbage collector was running on a fixed timer, so it would kick in mid-turn and steal CPU right when Claude Code was busiest. Now it waits until the process is idle.

[Sam Lambert @samlambert]

GitHub finally moved to serverless.

[Ivan Velichko @iximiuz]

August has been a weird month for iximiuz Labs so far. Signups and playground usage are up compared to June and July. But the revenue is significantly down - way below the sustainability level.

To stimulate things a bit, I'm running the Fair Pricing pilot - you can pick your discount, no questions asked. If you've been considering upgrading your membership, now is a good time to both support the labs and get a good deal: https://t.co/XEtsFHeIFc

Reposts are much appreciated, too!

And if you don't know what iximiuz Labs is, it's an indie learning-by-doing platform where you can master Linux, networking, containers, and Kubernetes in interactive VM-based playgrounds.