pkellner

Fable vs. Codex Review Notes

Question This Likely Answers

Why did Codex catch an issue that Fable missed? Does that mean Codex has a capability Fable lacks, and should I add a Codex review pass to the workflow?

On the Fable-vs-Codex Question

Codex's catch came from a question I had not explicitly asked:

pkellner

Add a "recognize flailing" guideline to AGENTS.md

Context

When the AI can't find a working solution it tends to thrash: trying variation after variation of a fix without converging. The operator wants a standing guideline that makes the AI recognize this pattern in its own responses, flag it, allow itself a bounded number of further attempts, and — if still stuck — stop and prompt the operator to investigate and possibly rethink the design, instead of burning time in a try-something-else loop.

pkellner

Presenting your own code changes

How to answer when the operator asks to see, explain, review, or understand changes you made. This complements "Code reviews are specific" in AGENTS.md, which governs reviews of code handed to you; this doc governs walkthroughs of your own work.

Treat "what changed?" as agentic code inquiry

When the operator asks anything like:

pkellner

name	tone-plain-no-bragging
description	How the user wants responses written — plain, humble, no self-congratulation
metadata	node_type type originSessionId memory feedback c588c490-68e8-4940-afda-a6a57c8e6a3e

The user dislikes self-congratulatory / bragging tone (e.g., "strictly better," comparison tables showing how clever a solution is, framing my own answers as smart). Be plain and humble, especially after mistakes.

pkellner

---
name: tone-plain-no-bragging
description: "How the user wants responses written — plain, humble, no self-congratulation"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: c588c490-68e8-4940-afda-a6a57c8e6a3e
---

The user dislikes self-congratulatory / bragging tone (e.g., "strictly better," comparison tables showing how clever a solution is, framing my own answers as smart). Be plain and humble, especially after mistakes.

pkellner

Here’s a formatted Markdown version you can paste into a GitHub Gist:

Why Cursor Runs Out Faster Than Claude Code on the Same Opus

You use Opus everywhere, but the two products are not metering the same work the same way.

Cursor exhausts faster for structural reasons, not because Opus behaves differently.

Same Opus model. Different meter. Cursor bills like prepaid API usage. Claude Code bills like a subscription with rolling resets.

pkellner

Here’s a formatted Markdown version you can paste into a GitHub Gist:

# Why Cursor Runs Out Faster Than Claude Code on the Same Opus

You use **Opus everywhere**, but the two products are **not metering the same work the same way**.

Cursor exhausts faster for structural reasons, not because Opus behaves differently.

> Same Opus model. Different meter.
> Cursor bills like prepaid API usage. Claude Code bills like a subscription with rolling resets.

pkellner

Your options represent different approaches to rendering and server-side involvement in a web application. Here's a breakdown of each approach:

1. CSR (Client-Side Rendering), no SSR

• How it works: The browser downloads a JavaScript bundle, which is then executed on the client to render the UI.
• Use case: Ideal for highly interactive, single-page applications (SPAs) where SEO is not a priority.
• Pros:
  • Fast initial server response (just static HTML/JS).
  • Easier to scale as it reduces server-side load.
• Cons:
• Slower Time to Interactive (TTI) because rendering waits for JavaScript to load.

pkellner

#!/bin/sh

# Define group prefixes
groups="better-rules-com better-rules-net gmail-to-icloud-net coolify"
group_totals="0 0 0 0"
unassigned_total=0

# Create temporary files
stats_file=$(mktemp)
temp_file=$(mktemp)

pkellner

#!/bin/bash

# Capture the stats for all containers
docker stats --no-stream --format "{{.Container}} {{.Name}} {{.CPUPerc}} {{.MemUsage}}" > stats.txt

# Temporary file to store container info for sorting
temp_file=$(mktemp)

# Helper function to format running time
format_time() {