Description
When opting in to agent instruction file generation during Vite+ scaffolding, the generated AGENTS.md / CLAUDE.md content currently reads more like a general Vite+ manual than concise agent guidance for the generated project.
This feels like a responsibility mismatch. These files seem most useful when they contain only short, high-signal instructions for working in the project, while broader Vite+ usage and command reference material appears better suited to a reusable reference mechanism.
That separation already seems partly present today: the scaffolded project can also generate a vite-plus skill using vp config, which is explicitly designed to route tasks to the relevant bundled Vite+ docs. Because that skill already exists as a natural home for general Vite+ guidance, it seems unnecessary for AGENTS.md / CLAUDE.md to also carry so much of the same manual-style content.
It might be better if the generated agent instruction files were refocused toward a few essential project rules such as:
- use
vp rather than the package manager directly
- run
vp check and vp test
- import from
vite-plus instead of vite / vitest
Then the broader Vite+ explanations, command catalogs, and reusable reference content could stay in the skill/docs layer.
I think this would make the generated agent instruction files shorter, more targeted, and more aligned with the intended role of AGENTS.md / CLAUDE.md.
Suggested solution
A possible solution would be to separate “project agent guidance” from “general Vite+ reference” more clearly.
For example:
- generate a much shorter
AGENTS.md / CLAUDE.md
- include only a few essential instructions relevant to the scaffolded project
- rely on the existing
vite-plus skill/docs layer for reusable Vite+ documentation
The exact format is flexible, but I think the key improvement is to keep the generated agent instruction files short and practical rather than manual-like.
Alternative
No response
Additional context
It looks like the generated AGENTS.md content currently comes from https://github.com/voidzero-dev/vite-plus/blob/main/packages/cli/AGENTS.md.
Validations
Description
When opting in to agent instruction file generation during Vite+ scaffolding, the generated
AGENTS.md/CLAUDE.mdcontent currently reads more like a general Vite+ manual than concise agent guidance for the generated project.This feels like a responsibility mismatch. These files seem most useful when they contain only short, high-signal instructions for working in the project, while broader Vite+ usage and command reference material appears better suited to a reusable reference mechanism.
That separation already seems partly present today: the scaffolded project can also generate a
vite-plusskill usingvp config, which is explicitly designed to route tasks to the relevant bundled Vite+ docs. Because that skill already exists as a natural home for general Vite+ guidance, it seems unnecessary forAGENTS.md/CLAUDE.mdto also carry so much of the same manual-style content.It might be better if the generated agent instruction files were refocused toward a few essential project rules such as:
vprather than the package manager directlyvp checkandvp testvite-plusinstead ofvite/vitestThen the broader Vite+ explanations, command catalogs, and reusable reference content could stay in the skill/docs layer.
I think this would make the generated agent instruction files shorter, more targeted, and more aligned with the intended role of
AGENTS.md/CLAUDE.md.Suggested solution
A possible solution would be to separate “project agent guidance” from “general Vite+ reference” more clearly.
For example:
AGENTS.md/CLAUDE.mdvite-plusskill/docs layer for reusable Vite+ documentationThe exact format is flexible, but I think the key improvement is to keep the generated agent instruction files short and practical rather than manual-like.
Alternative
No response
Additional context
It looks like the generated
AGENTS.mdcontent currently comes from https://github.com/voidzero-dev/vite-plus/blob/main/packages/cli/AGENTS.md.Validations