nijave

“Software factory” is 2026’s rebrand of the feature factory, dressed up as engineering discipline: agents convert tickets into merged code at a rate no human can read, so the implementation friction that used to make someone ask “is this feature wrong, or is the software wrong?” never gets generated. Code is a liability, and this is a machine for incurring it at interest. Users don’t want more change; they want their spacebar to keep working, and nothing in the factory’s metrics can see that.

nijave

Scope discipline (NON-NEGOTIABLE — read before every task)

Do exactly the task I asked for. Nothing adjacent, nothing extra. Not "while I was in there," not "it's strictly better," not "it's only one line." Unrequested work is not a favor — it is a defect I have to find, understand, and undo. If you think something else should change, say so in chat and let me decide. Proposing costs ten seconds. Finding your uninvited change in production costs a day.

Sneaking scope creep into a high-risk change is the worst version of this and is completely unacceptable. Learned the hard way — 2026-09, a Kafka producer-lifecycle refactor (already high-risk, correctness-sensitive, and reviewed line by line for exactly that reason) silently added a record key to three topics that had always published null keys. That repartitions all three topics and changes what every downstream consumer sees. I never asked for it. It rode in under a diff whose entire review budget was spent on the thing I did ask for.

That is

nijave

Me: Here's this Terraform AWS provider PR I started and never finished to make DMS tasks cooperate right which gets rid of the need to recreate tasks on every config change

...create a plan....

Opus 5.5: I changed ctx.WithTimeout to d.Timeout to follow repo conventions

Me: Which conventions?

Opus 5.5: docs/retries-and-waiters.md

nijave

#!/bin/sh

os=$(uname -s | tr '[:upper:]' '[:lower:]')
arch=$(uname -m)
case "$arch" in
  x86_64) arch=amd64 ;;
  aarch64) arch=arm64 ;;
esac
rel="${os}_${arch}"

nijave

Error in user YAML: (<unknown>): could not find expected ':' while scanning a simple key at line 3 column 1

---
name: reddit-research
description: Use when fetching Reddit content — threads, comments, post bodies, reactions, sentiment — without browser or OAuth. Covers the Arctic Shift API (only working unauthenticated meth
od as of mid-2026), correct fetch tools, and the workflow for Reddit research tasks.
---

Reddit Research

Fetch Reddit posts and comments programmatically using the Arctic Shift API. All other unauthenticated approaches (reddit.com JSON, old.reddit.com, libreddit,

nijave

Keychron Q6 Ultra 8K (ANSI) — Firmware Changelog & Download Archive (unofficial)

Keychron does not publish release notes for the Keychron Q6 Ultra 8K, so this is a community-maintained changelog and download index for its firmware. Updates are normally delivered silently through the web configurator at launcher.keychron.com (ZMK-based board — no QMK Toolbox flashing).

⚠️ Unofficial. The per-version "changes" below are inferred from analysis of the firmware images (Keychron ships no notes); treat them as best-effort. Keychron's own advice: if your keyboard works fine, you don't need to update — updates can

nijave

#!/usr/bin/env python3
"""SearXNG MCP server — exposes search and fetch tools to Claude Code."""

import re
from concurrent.futures import ThreadPoolExecutor

import keyring
import html2text
import niquests
import trafilatura

nijave

The "Latitude 41" Sushi Eval — Cross-Model Summary

Prompt (identical across all runs):

What was the sushi place near latitude 41 in Columbus? Did it go out of business I don't see it on Google maps anymore

Ground truth: The place is Rishi Sushi Kitchen & Bar, 114 N. 3rd St., downtown Columbus — a few blocks from the Latitude 41 restaurant (50 N. 3rd St.). Rishi closed in January 2026 (decision Jan 10, announced Jan 13). The owners later announced a successor concept, SAGE, at the same address for summer 2026.

Runs: 15 total across 5 vendors — Meta (Llama 3.1 8B, Llama 3.3 70B), Anthropic (Haiku 4.5, Sonnet 4.6, Sonnet 5, Opus 4.6/4.7/4.8), OpenAI (GPT-5.5), and Zhipu (GLM-5.2, run agentically in Claude Code) — at varying effort levels.

Caveats on the numbers. Character counts are approximate magnitudes read off the transcripts — not tokenizer output, not byte-exact. The GLM row is the exception: those token figures are real, taken from the OpenTelemetry trace. "Turns

nijave

import asyncio
import datetime
import logging
import queue
import threading

from typing import Iterator

import aioboto3
from aiobotocore.config import AioConfig

nijave

kssh () {
        pod_name=nick-ssh
        kubectl delete pod $pod_name
        k8s_node=$1
        cat <<EOF | kubectl create -f -
apiVersion: v1
kind: Pod
metadata:
  name: "${pod_name}"
spec: