MoonBit Pi extension

Wraps moon ide subcommands (plus check, test, and the fmt+info handoff) as discrete Pi tools, so the model reaches for compiler-aware navigation instead of grepping .mbt source.

https://www.npmjs.com/package/pi-moonbit

Install

One of

pi install npm:pi-moonbit

pi install git:[email protected]:kaashyapan/pi-moonbit.git

pi install https://github.com/kaashyapan/pi-moonbit.git

Tools registered

Navigation / IDE

• moon_peek_def — resolve a definition, with source context
• moon_find_references — find usages across dependents
• moon_type_info — type and docs at a position (wraps moon ide hover)
• moon_outline — structure of a file or package (positional path)
• moon_rename — compute rename edits (dry run unless apply: true)
• moon_analyze — public API usage counts
• moon_doc — search exported APIs/docs
• moon_error_explain - Explain an error code
• moon_workspace_symbols - Search top-level symbols (functions, types, enums, traits, methods) across every package

Build / quality

• moon_check — static analysis via moon check --output-json, summarized
• moon_test — run tests (-p, --target, optional --update)
• moon_fmt_info — moon fmt then moon info (handoff sequence)

Caveat

• The extension defaults to run its tools on the wasm-gc target for speed.
• The LLM can override with the supported_target.
• If the LLM fails to do that, the tools may not work.
• If you encounter problems raise an issue

Why these tools - Token efficiency

1. Calling dedicated tools that return json instead of streaming to and from bash is more token efficient.
2. Keeping in pi spirit with minimal prompting, the package includes a tersely written system prompt describing the tools.
3. Some QOL improvements, like moon check does not return diagnostic messages from dependencies.
4. The error explain tool will placed to save a lot of tokens, where the model only needs to look up the error code necessary instead of running an explain on the entire project.
5. Better tool reliability. Higher chances that the model will use the tool for the purpose. Higher chances of the model picking moon ide over grep on the entire codebase. Verify with tool-stats.

Important

• The package injects a system prompt that includes the necessary instructions to use the tool chain
• It is important that your skills and AGENTS.md and other prompt files DO NOT mention moon check, moon ide, moon test etc.. It will only confuse the model. Refer to this project's AGENTS.md for a sample.
• The package blocks running these commands over bash. Giving these instructions will most likely confuse the model.