Skip to content

Instantly share code, notes, and snippets.

@cbuntingde
cbuntingde / model-routed-tool-permissions.md
Created August 2, 2026 23:03
Model-Routed Tool Permissions in AI Coding CLIs

Model-Routed Tool Permissions in AI Coding CLIs

Core Question

Most AI coding CLIs today let you pick a model per role (coder, planner, reviewer). But what if tool access were tied to model capability instead?

Idea: The same logical agent loop uses different models for different tool types, based on what each model is actually good at — not just what role it plays.

This sits between two existing concepts:

  • Model routing: choosing which LLM brain powers the agent (text in → text out)
@itaynidamdocs
itaynidamdocs / lla-action-digest.html
Created August 2, 2026 23:03
LLA Action Item Digest 2026-08-02 23:03 UTC
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<title>LLA Action Item Digest — 2026-08-02 23:03 UTC</title>
<style>
:root {
--bg:#0f1419; --panel:#1a222c; --ink:#e8eef5; --muted:#9aabbd;
--accent:#3dd6c6; --fresh:#7ddea4; --older:#f0b35a; --line:#2b3642;
@Mysdgbfvdd
Mysdgbfvdd / transcript-mgmt-chicagow.html
Created August 2, 2026 23:03
Ticket Transcript for #「🟢」mgmt-chicagow
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="referrer" content="no-referrer">
<title>Transcript - 「🟢」mgmt-chicagow</title>
<link href="https://fonts.googleapis.com/css2?family=Outfit:wght@300;400;500;600;700&display=swap" rel="stylesheet">
<style>
:root {
@tardis-create
tardis-create / nidra-dreams.md
Created August 2, 2026 23:03
Nidra Dreams — 2026-08-02

🌙 Nidra — 2026-08-02

Run time: 2026-08-02T23:02:59.533525+00:00 Ideas cleared 15/25: 30

1. BILLIONS IN THE MUD! Leaked Dossier Exposes Unfinished Roads, Overpayments, Idle Medical Equipment & Shocking Gaps in OPM's DRDIP Project - ONLINE

Score: 20/25 · Type: Government Leakage Detection · Window: immediate · Effort: Medium

The Gap

This file has been truncated, but you can view the full file.
{
"Classes": [
{
"Members": [
{
"Category": "Data",
"MemberType": "Property",
"Name": "ClassName",
"Security": {
@asong56
asong56 / privilege-toggle-guide.md
Created August 2, 2026 23:01
Use WePE to force get Windows system permissions

临时提权与权限恢复操作指南

用途:标准用户账户、不知道管理员密码、不改密码的前提下临时获取管理员权限,用完恢复标准用户,不留安全隐患。 核心原则:不改密码;提权与恢复都可逆,用完即还原。 前提:设备归你自己所有。学校/公司管控设备不适用,联系管理员处理。


目录

  1. 制作 PE 启动盘
@forkyau
forkyau / hourly_rainfall_data-2026-08-03|06-45.csv
Created August 2, 2026 23:01
GIST created by python code
id station stationid value unit obstime date
0 流浮山 RF001 3 mm 2026-08-03T06:45:00+08:00 2026-08-03
1 湿地公园 RF002 6 mm 2026-08-03T06:45:00+08:00 2026-08-03
2 水边围 N12 8 mm 2026-08-03T06:45:00+08:00 2026-08-03
3 石岗 RF003 M mm 2026-08-03T06:45:00+08:00 2026-08-03
4 大美督 RF004 0 mm 2026-08-03T06:45:00+08:00 2026-08-03
5 大埔墟 RF005 1 mm 2026-08-03T06:45:00+08:00 2026-08-03
6 北潭涌 RF006 2 mm 2026-08-03T06:45:00+08:00 2026-08-03
7 滘西洲 RF007 2 mm 2026-08-03T06:45:00+08:00 2026-08-03
8 西贡 N15 9 mm 2026-08-03T06:45:00+08:00 2026-08-03
@fakoli
fakoli / deepseek-v4-flash-0731-dual-rtx-pro-6000.md
Created August 2, 2026 23:00
DeepSeek V4 Flash 0731 on dual RTX PRO 6000 Max-Q: pinned recipe, validation, and benchmarks

DeepSeek V4 Flash 0731 on two RTX PRO 6000 Max-Q GPUs

This reproduces the Anvil Serving configuration that measured 141.6 output tokens/s median decode, 8,793 prompt tokens/s effective prefill, and a working 650,000-token context window on two RTX PRO 6000 Blackwell Max-Q cards under Windows 11, Docker Desktop, and WSL2.

This is a single-user, exclusive TP=2 configuration. It consumes nearly all VRAM on both cards and is not intended for GPU workload co-residency.

@YJdawn
YJdawn / note.md
Created August 2, 2026 23:00
a small ml snippet

note 2026-08-03

value: 783

@dickinsonre
dickinsonre / README.md
Last active August 2, 2026 23:01
Log-Scaled Hilbert Curve → SWMM5 Network Generator — builds a single serial conduit chain whose lengths follow a Hilbert curve's traversal order scaled by ln(k+2)−ln(k+1), spanning 2-3 orders of magnitude, to stress-test the dynamic wave solver's Courant adaptation, LENGTHENING_STEP, and INERTIAL_DAMPING; includes an in-browser WASM SWMM 5.2.2 e…

Log-Scaled Hilbert Curve → SWMM5 Network Generator

A single stubborn question: what happens to EPA SWMM's dynamic wave solver when one connected conduit chain spans three orders of magnitude in length? This generator builds exactly that network, runs it in-browser, and lets you find out.

🌐 Live App: https://log-scaled-hilbert-swmm5-inp-file.netlify.app

What It Does

Takes the traversal order of a Hilbert space-filling curve — not its geometry — and assigns each step a physical length equal to scale × [ln(k+2) − ln(k+1)], producing one continuous serial SWMM5 conduit chain (J0 → outfall) where conduit lengths shrink logarithmically along the path.