@xl0/pi-lovely-codex

Pi extension for GPT fast mode and Codex-style apply_patch

Packages

Package details

extension

Install @xl0/pi-lovely-codex from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:@xl0/pi-lovely-codex
Package
@xl0/pi-lovely-codex
Version
0.2.5
Published
Oct 23, 2026
Downloads
195/mo · 60/wk
Author
xl0
License
MIT
Types
extension
Size
28.1 KB
Dependencies
1 dependency · 4 peers
Pi manifest JSON
{
  "extensions": [
    "./extensions"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

@xl0/pi-lovely-codex

GPT "Fast mode" control and Codex-style apply_patch tool.

Install

pi install npm:@xl0/pi-lovely-codex

Or use without installing:

pi -e npm:@xl0/pi-lovely-codex

/lovely-codex

Lovely codex settings

GPT fast mode 🏎️

Sends service_tier: "priority" to request priority serving:

  • default — don't send service_tier
  • fast — priority on both openai and openai-codex
  • fast-codex — priority on openai-codex only

GPT models on those two providers only. Fast mode shows 🏎️ in the status line.

Availability and billing depend on your account and model. Codex responses can report service_tier: "default" even when priority measurably increases throughput; that field alone does not tell you whether Fast mode worked.

apply_patch

GPT models are trained on Codex's apply_patch; with Pi's edit tool, muscle memory shows — silly mistakes creep in. This extension gives them the tool they were trained on.

By default, apply_patch is a plain JSON tool { input: string }. For models that support OpenAI grammar tools, it can instead be a freeform custom tool constrained by Codex's Lark grammar.

Patches are applied by shelling out to codex --codex-run-as-apply-patch. The Codex CLI must be installed and on PATH.

Settings:

  • add apply_patch — on, off, or gpt-only (default: only for model ids starting with gpt- or containing /gpt-).
  • enable freeform — use the grammar-constrained custom tool when supported (off by default).
  • disable write / disable edit — drop the now-redundant built-in tools while apply_patch is active (off by default); restored after, but never beyond what the session started with.
  • disable read — on, off, or gpt-only (default: only for GPT model ids, independent of apply_patch). Forces bounded shell reads (sed -n/ head) instead of whole-file reads, conserving context the way Codex's toolset does. While read is gone, a view_image tool takes over image reading (which the shell can't provide); turn it off with add view_image.

Config files

  • User: ~/.pi/agent/xl0-pi-lovely-codex.json
  • Workspace: <cwd>/.pi/xl0-pi-lovely-codex.json

Workspace overrides User. All keys optional.

Related projects
Pi Lovely Web web_search, web_fetch, web_image tools
Pi Lovely Dev Tools interactive debugging helpers /tool, /show-sysprompt, /show-context, /llm-stats
Pi Lovely IDE IDE integration
Pi Lovely Config scoped config helpers for Pi extensions
Pi Lovely Comment open the last assistant message in your editor and sync edits back into the prompt
Pi Lovely Rename automatic and manual session naming

Like this work? Hire me