You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Instantly share code, notes, and snippets.
⚽
I didn't drink lots of milk, and I still wasn't good enuf for Accrington Stanley
Mike Preston
Darkflib
⚽
I didn't drink lots of milk, and I still wasn't good enuf for Accrington Stanley
Agentic infrastructure · SRE · MCP servers · UK · Available for fixed-scope reviews
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
<title type="text">The kitchen of the future is here — and it is steam powered</title>
<summary type="text">Our interiors editor reports from the cutting edge of appliance design, including steam drawers, steam cabinets and — yes — dishwashers that steam your plates</summary>
OpenSpec: A Deep Primer on Spec-Driven Development for AI Coding Assistants
OpenSpec is a lightweight spec-driven development (SDD) framework designed specifically to align humans and AI coding assistants before any code gets written. Think of it as "version control for intent" - you create explicit, reviewable specifications that guide AI assistants to build what you actually want, not what they infer from chat history.
Core Philosophy: Brownfield-First & Single Source of Truth
OpenSpec distinguishes itself from alternatives (GitHub's Spec Kit, AWS Kiro) through two key principles:
1. Brownfield-First Design
Most SDD tools assume greenfield (0→1) development. OpenSpec excels at existing codebase modification (1→n), especially when changes span multiple specifications. The folder structure reflects this:
From a conversation with GPT late at night some months ago:
My mind back when you mentioned it jumped towards the common argument that 'guns are just tools' and 'tools are neutral'. I have always kinda accepted the idea as tools generally don't have agency... But in the modern world this is becoming more fuzzy.
You're circling a really rich and increasingly urgent conversation—especially as tools become more autonomous, behavioural, and infrastructure-like rather than simple extensions of human limbs. Let's unpack this from several lenses:
Here is a simplified, accessible summary of the FIPS (Federal Information Processing Standards) regime, and a sketch of other U.S. security / communications rules you should know — plus caveats. (If you like, I can also show how these compare with UK/EU equivalents.)
What are FIPS / the FIPS “data-processing guidelines”?
FIPS are a set of standards published by NIST (the U.S. National Institute of Standards and Technology) for use by U.S. federal agencies — and often by their contractors and vendors — especially when handling sensitive information. ([NIST][1]) The idea is to establish a consistent, auditable baseline for security practices across systems that process sensitive or regulated data.
Here are key elements and rules in FIPS that relate to data processing and communications:
Here’s a one-off “morning update” test run (Europe/London). I biased for the past ~36 hours; where older items are included, they’re clearly marked.
#1 — Anthropic status today: elevated errors on Opus 4 and Sonnet 3.5; earlier Sonnet issues on 25 Sep were resolved. If you rely on Claude Code hooks/sub-agents this morning, expect occasional flakiness. ([status.anthropic.com][1])
#2 — PyPI phishing targeting Python developers continues. Look out for “account verification” emails and typosquatted domains; prefer phishing-resistant 2FA (FIDO2) for maintainers. ([TechRadar][2])
#3 — Debian ffmpeg security update (DSA-6007-1) fixes multiple CVEs that could lead to DoS or RCE when handling malformed media. Patch homelab encoders/players if you do HLS/RTSP. (Published 6 days ago, still high-impact.) ([Debian][3])
#4 — Google Cloud release notes (24 Sep): Cloud Run .env file support is GA, and a carbon-footprint data correction for europe-west2 (London) was issued; re-export Jul
Python preferred over node.
Include class and function doc blocks.
Default license is MIT.
Use SemVar.
Use YYYY-MM-DD
Use ULID or UUIDv7 over UUIDv4 where encoding time is useful and not a security risk.
Define JSON formats using JSON Schema.
Diagrams in mermaid or PlantUML format - bias toward mermaid.