What is pi-skill-model?

pi-skill-model is a small extension for Pi, an AI coding assistant, that lets you pin a specific AI model to a specific skill. Normally Pi runs every skill on whatever model the current session is using, but this extension adds a model field to a skill's frontmatter so you can force that one skill to always run on a particular model, no matter what the rest of the session is set to. For example, a project might have a skill for writing a product requirements document, and the author can set model to something like anthropic/claude-opus-4-8 in that skill's SKILL.md file. From then on, running that skill always calls Opus, even if the user is chatting with a cheaper or different model elsewhere in the session. You can point to a model either with an exact provider and id, such as zai/glm-5.2, or with a looser text pattern like opus, though the exact form is recommended for skills you plan to share since loose patterns can break when model names change later. One deliberate design choice is that skills have no fallback: if the specified model is not available, the skill simply will not run, rather than quietly switching to a weaker or pricier substitute. The extension also checks that every model referenced in your skills is actually available whenever Pi starts up or when you run its reload command. Subagents are not affected by this extension, only skills. Installation is a single command, pi install git:github.com/MitchelliJ/pi-skill-model, which can be pinned to a specific tag or run with a flag to install at the project level instead of user level. After installing, Pi needs to be restarted or reloaded for the change to take effect. The project is released under the MIT license.