Conversation
Reframes the repo as a general collection of AI coding agent skills rather than pi-specific. Skills are now documented as standalone markdown files usable with Claude Code, Codex, Cursor, Windsurf, pi, or any agent that loads markdown context. - README: new Setup section with agent-specific wiring instructions (slash commands, skills directories, @-imports, direct reference, pi) - README: relative links on all skill and extension entries - README: extensions clearly labelled as pi-only - multi-review: remove pi CLI invocation and broken code-review skill reference; make sub-agent step tool-agnostic; replace Opus/Codex labels with Model A/Model B throughout - package.json: update description and keywords Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
The package.json is pi's package manifest — the keywords and description are consumed by pi tooling, not for general discovery. Shouldn't have touched it. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Add full install options (local path, GitHub URL, manual settings.json), explain what pi does with the package once installed (skills as slash commands, extensions as hooks/tools), and clarify the update workflow. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
pi is fundamentally different — it uses a package manager and loads both skills and extensions. Split Setup into two distinct top-level sections: - "Setup — other agents (skills only)" for symlink/reference approaches - "Setup — pi (skills + extensions)" for the package install workflow Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
There was a problem hiding this comment.
Couple thoughts on this:
- Leaving the models and command to run the sub-agents unspecified will make the skill less token efficient as the model has to figure out how to do this every time, and also potentially makes each run less consistent. I get wanting the idea is to make it not pi-specific, but it makes it less efficient when we do that.
- Maybe it makes more sense to make a more minimal change to this skill that says something along the lines of "If these two models are not available then choose the next most appropriate ones from the models that are available"
There was a problem hiding this comment.
While skills are largely portable, extensions (at least as they exist in this repo) are very much a pi-specific thing since they hook into pi's extensions API.
There was a problem hiding this comment.
Yes correct. This PR doesn't change that; actually better explains it in the readme.md no?
nishantdesai
left a comment
There was a problem hiding this comment.
Thanks for updating the details to reflect what the repo is actually about. Skills are after all agent agnostic.
pweiskircher
left a comment
There was a problem hiding this comment.
I don't care overly much either way. I thought this would be good so we can use the pi isntall of a gitrepo and have extensions and stuff but I can deal with whatever!
Reframes the repo as a general collection of AI coding agent skills rather than something pi-specific. Skills are standalone markdown files that work with Claude Code, Codex, Cursor, Windsurf, pi, or any agent that loads markdown context.
README
Setup — Skills (with Claude, Codex, etc)section covering slash commands (Claude Code, Cursor, Windsurf), skills directories (Codex's.agents/skills/, Claude Code's.claude/skills/),@-imports inCLAUDE.md, and direct prompt referenceSetup — pi (skills + extensions)section with full install options (local path, GitHub URL, manualsettings.json), explanation of what pi loads, update workflow, and links to the pi website and repositorymulti-review
pi -pCLI invocationcode-reviewskillOpus/Codexlabels withModel A/Model Bthroughout