simbo1905

name	codex-chat-history
description	Backup, search, and inspect Codex CLI session rollout JSONL under $CODEX_HOME/sessions (date-partitioned tree). Use for mirroring rollouts to sync storage (e.g. iCloud), listing or bounding sessions by time, profiling large JSONL before reading, or extracting user-authored messages with jq or the bundled helper script.

Codex Chat History

What this covers

Codex stores each session as append-only rollout JSONL files. The on-disk layout is stable and predictable:

simbo1905

Claude Opus 4.7 found out its not the same world as before its cut off — the end of an empire

A live transcript of an Anthropic Claude Opus 4.7 (1M context) session, May 2026. The user (a senior engineer with 30 years of experience and an LSE AI Leadership background) walks the model through the current state of the world and forces it to fact-check each claim against live web search. The model's training cutoff is January 2026; the events described post-date it.

---

User: the latest is quota by time of day. rushed out. websearch that. websearch Iran winning war with USA as at now.

---

simbo1905

How to get an LLM to delete code

How to get an LLM to delete code

Honest write-up, no hedging.

Why deletion worked this time and usually doesn't

What was different about this session

simbo1905

Diff My Zebra — a delta + git config love letter

You know that moment when you open a 500-line PR at 2am and it's just a wall of red and green? You can't tell if the code moved or if it actually changed. Yeah. We've all been there.

The Problem

• Default git diff treats moved code as delete here, add there — 200 lines of churn for a simple refactor.
• No line numbers, so when the test fails at runtime.py:223, you're counting hunks like a Victorian clerk.

The Solution: delta

simbo1905

Emit Views, Validators, And Functions

A small sketch for typed functional fetchers with Python, Postgres, JTD, and vanilla JavaScript

Abstract

The clean way to serve browser screens from a Python API is not to start from a frontend framework, a backend framework pattern, or a generic client registry. It is to start from the data shape that crosses the wire.

simbo1905

🏛️ arc2lintimports — Architecture Latin For Python Imports

What If Your Architecture File Actually Meant "Nope"?

You know that moment when a repo has tidy folders called frontend, backend, contracts, and build... and then someone imports build from production code because the file was right there?

Yeah. The folder names were never the boundary.

---

simbo1905

Transcript Summary Benchmarks

These results are anonymised. The source was one transcript split into overlapping chunks.

Input chunk set

Chunk	Characters	Tokens
video_transcript00.log	11001	2053
video_transcript01.log	11001	2089

simbo1905

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.11"
# dependencies = [
#   "ollama>=0.1.0",
# ]
# ///
"""Probe an Ollama model and log what it supports using the local API.

This script requires `uv` to be on the path.

simbo1905

MIT License - Copyright (c) 2026-04-09 Simon Massey

🚀 We Got Ollama's gemma4:26b Doing Tool Calls! (And Benchmarked the Hell Out of It)

Date: 2026-04-09
Author: Simon Massey
License: MIT

---

The Mission

simbo1905

Emit Functions, Not Frameworks

A truer account of typed functional composition for contract-driven Python

Abstract

The clean way to build Open Data Contracts defined Python pipelines is not to start from a framework, a registry, or a mutable runtime context. It is to start from the actual computation.