Development

Repo layout

Path	What it is
`skills/`	Skill definitions (`SKILL.md`) that back each slash command
`commands/`	Slash-command prompts (`.toml`/`.md`) — `margaret`, `margaret-scan`, `margaret-review`, `margaret-guard`, `margaret-design`, `margaret-help`
`hooks/`	Lifecycle hook scripts, config resolver, statusline scripts, uninstaller
`.claude-plugin/`, `.codex-plugin/`	Host-specific plugin manifests
`.cursor/`, `.windsurf/`, `.clinerules/`	Static rules-file copies for hosts with no hook system
`AGENTS.md`	Always-on context file for Gemini CLI and generic agent hosts
`benchmarks/`	Response-quality benchmarks comparing margaret against baseline/caveman arms
`tests/`	Unit tests for hook logic (config, instructions, mode tracker, runtime, statusline, uninstall)

Running tests

npm test

This runs the unit test suite, checks that every command/skill declared in plugin.yaml is backed by a real file (scripts/check-structure.js), and checks that the persona’s static copies (AGENTS.md, .cursor/, .windsurf/, .clinerules/) haven’t drifted from each other (scripts/check-rule-copies.js).

Keeping persona copies in sync

The screen text is duplicated across AGENTS.md, .cursor/rules/margaret.mdc, .windsurf/rules/margaret.md, and .clinerules/margaret.md for hosts without a hook system. Edit skills/margaret/SKILL.md first, then propagate the change to each static copy — npm test fails the build if they drift.

Benchmarks

benchmarks/ scores margaret’s output against a plain baseline and the caveman skill on correctness and lines-of-code, via promptfooconfig.yaml and the prompts in prompts.json. See benchmarks/README.md for how to run an eval pass.

Getting Started

Reference

Contributing

Repo layout

Running tests

Keeping persona copies in sync

Benchmarks

License

​Repo layout

​Running tests

​Keeping persona copies in sync

​Benchmarks

​License

Repo layout

Running tests

Keeping persona copies in sync

Benchmarks

License