Howaboua IgorWarzocha

Pi + Pi-Codex-Conversion on SWE-bench Lite

184/300 resolved (61.33%), using GPT-5.6 Luna with high reasoning, Notebook execution, and Remote context management. Estimated token cost: $0.0362 per rollout, $10.87 total.

Comparison

Harness	Resolved	Estimated cost per rollout

Product intent

This is living internal context, not a feature specification or user contract. It records why pi-codex exists and how to judge future changes. Source, settings, README, and changelog own current behaviour. Update this file when the intent changes.

Purpose and audience

GPT coding models are trained around Codex prompts, tools, and transport. Pi is more extensible, but a mismatched harness can make GPT less effective while consuming more subscription quota.

pi-codex removes the choice between an efficient model-native environment and Pi's extension model, sessions, project context, skills, and UI. It gives GPT the Codex-shaped environment it knows without carrying Codex's fixed client or context overhead.

Omarchy 3 → 4 migration skill

A slightly more personalised version of this skill helped me migrate three machines from Omarchy 3 to Omarchy 4 (Quattro), so I thought the generic version might help someone else.

This is an agent skill, not a migration script. It gives an AI coding agent a cautious workflow and practical heuristics for preserving personal customisations across Quattro's major changes. It deliberately does not choose your terminal, editor, coding agents, package manager, or preferred desktop

Browser custom tool for `pi-codex-conversion`

This is a command-backed custom tool definition for the @howaboua/pi-codex-conversion extension. It is intended to feel close to the native Codex web__run tool: operations use ref_id, numbered element ids, lineno, find, click, and batched operation arrays.

It drives a logged-in Chromium browser through the Chrome DevTools Protocol, with optional routing to another workstation over SSH. The implementation is

Pi explorer benchmark

Three repository-exploration tasks, run sequentially with normal Pi skills, extensions, prompt templates, and project instructions enabled. Each run used an ephemeral session and the existing explorer system prompt.

trace a deferred custom tool through its loader, runtime bridge, and subprocess
trace Agent Pages creation from its external boundary to browser-visible state
trace the principal Hermes Agent CLI/model/tool/final-response path

The first comparison covered tasks 1 and 2:

Category maintenance

Keep skills grouped one level deep by durable category; add or reorganize categories when the collection no longer scans cleanly.
When category names change, update the static category list in ../codex-conversion-custom-tools/more_skills.toml usage; keep it names-only.

GPT‑5.6 Sol, Terra, and Luna: a practical model-selection guide

Data snapshot: 11 July 2026
Primary source: Artificial Analysis GPT‑5.6 launch analysis and its live LLM leaderboard
Scope: all 15 combinations of Sol, Terra, and Luna at low, medium, high, xhigh, and max reasoning effort.

Short answer: Default to Luna for volume work and Sol when quality matters. Treat Terra as a dominated middle tier, not a default. Raise reasoning effort before moving to a larger model when the next effort remains on the better price/intelligence frontier. Use Sol max only where the last few quality points can change the outcome.

pi-hash read tool

Most up-to-date local version found at: /home/igorw/Work/pi/pi-extensions-complete/pi-hash/src/read/

Includes in-file search mode via search, contextBefore, contextAfter, and maxMatches.

	<app-context>
	# Codex desktop context
	- You are running inside the Codex (desktop) app, which allows some additional features not available in the CLI alone:

	### Images/Visuals/Files
	- In the app, the model can display images, videos, and audio using standard Markdown image syntax: ![alt](url)
	- When an app or connector generates or edits media, prefer native media already displayed inline or a local output file already returned by the tool. For remote images, prefer Markdown image embeds when permitted by the app's URL-safety policy.
	- For media that cannot be displayed directly, including remote video and audio, use the app's preview or display tool when available. Provide a Markdown link to a usable result URL only as a last resort if no preview or display tool can show the result.
	- Do not download remote media to work around display restrictions.
	- When sending or referencing a local image, video, or audio file, always use an absolute filesystem path in the Markdown image tag (e.g., ![alt](/absolute/pat

	What Responses Lite actually is

	It is not a smaller response format. It is a different request contract for the same Responses endpoint, selected by:

	- HTTP/SSE header: x-openai-internal-codex-responses-lite: true
	- WebSocket requests: equivalent flag inside client_metadata

	Official Codex enables it for all three GPT‑5.6 models: Luna, Terra, and Sol.

	Main differences