Magnet CLI
Install:
pnpm i -g @heypal/magnet-cli@latest
How to run:
OPENAI_API_KEY=<key> magnet run
magnet run
How to run locally:
pnpm install
pnpm build && pnpm link --global
magnet run
This creates a .magnet/magnet.json
file in the current directory if it doesn't exist already.
Commands
magnet run
Runs the plugins specified in index.ts
passed to the run
command. So far, we have the following plugins:
- FilePlugin, which keeps the metadata
magnet.json
file up to date. - LlmPlugin, which generates summary information about a file by passing its contents to OpenAI and whether it's an API route or Webapp Page
- OpenApiPlugin, which matches on files where
llm.isApiRoute=true
, passes the file contents to OpenAI to generate an OpenAPI schema that's saved to openaiSchema.json
under the file's sub-directory in .magnet
magnet clean
Cleans up the .magnet/magnet.json
metadata file.
- Removes entries for which there is no corresponding file in the root directory (the directory from which
magnet clean
was called - Removes entries which match any of the
ignorePatterns
- Deletes deprecated fields
Plugins
Plugins get passed the full metadata
object, which is the entire magnet.json
file. They can then choose what to operate on themselves and update the magnet.json
file arbitrarily.
Most of the time, the plugin will read files
, which should be a map of file paths to file metadata, generated by the built-in FilePlugin
. Other plugins are expected to add their output to files[<filepath>][<pluginKey>]
, so it doesn't conflict with the metadata generated by other plugins.
For example, the init.ts
file has the following entry. Notice the LLM Plugin saves its output inside the llm
key.
{
"files": {
"/Users/dev/magnet-cli/commands/init.ts": {
"notesPath": "/Users/dev/magnet-cli/.magnet/notes/magnet-cli/commands/init.ts",
"lastChanged": "2023-08-18T16:05:55.211Z",
"path": "/Users/dev/magnet-cli/commands/init.ts",
"llm": {
"lastAnalyzed": "2023-08-21T00:07:01.257Z",
"needsAnalysis": false,
"output": "{\n \"summary\": \"This file is responsible for initializing the application. It has an 'init' method which is used to log the directory root of the application context.\",\n \"keywords\": [\"import\", \"zod\", \"BaseCtx\", \"InitCtx\", \"async function\", \"console.log\"],\n \"description\": \"This script is used to initialize an app using a provided context.\",\n \"isApiRoute\": false,\n \"apiRouteDescription\": null,\n \"isWebappPage\": false,\n \"webAppPageDescription\": null\n}",
"summary": "This file is responsible for initializing the application. It has an 'init' method which is used to log the directory root of the application context.",
"description": "This script is used to initialize an app using a provided context.",
"keywords": [
"import",
"zod",
"BaseCtx",
"InitCtx",
"async function",
"console.log"
],
"isApiRoute": false,
"apiRouteDescription": false
}
}
}
}
LLM Plugin
This only runs on files whose llm.lastAnalyzed
is null, or llm.needsAnalysis
is explicitly set to true
, or llm.lastAnalyzed
is further in the past than lastChanged
.
OpenAI Plugin
This only runs on files whose llm.isApiRoute=true
.
Development
New Github releases are published in the publish.yaml
workflow. These releases are created when new semver tags are pushed. See the release.yaml
workflow.
To bump the version and push a release, run pnpm release:patch
.
This builds the project, runs tests, bumps the version number and creates a new commit for the new version, and creates then pushes a new tag for the version number to Github.
It is suggested to title commits according to Conventional Commits, so the release workflow will pick up changes and put them in the changelog.
This means prefixing titles with feat:
or fix:
, e.g. fix: properly handle timeout errors
.
See the full list of prefixes in the changelogithub.js
file.
If a commit title isn't prefixed, it will not be picked up.
Example OpenAPI schemas
Additional notes
See DEVELOPMENT.md