Claude Code plugin
The MinLang Claude Code plugin turns Claude Code into a MinLang-aware authoring, migration, refactoring, and release tool. It ships a namespaced set of slash commands (/minlang:*), optional sub-agents, a non-destructive edit hook, and an optional local ml1 MCP bridge — all wired to load the canonical language bundle before producing or validating any MinLang.
The plugin does not bundle the compiler. It drives the ml1 CLI you install separately (see Getting started and the CLI reference) and treats real ml1/make runs as the authoritative verification step.
- 23 slash commands across five families — see Commands.
- 3 optional agents, a shared skill, an edit hook, and an MCP bridge — see Agents & tooling.
- Bundle authority discipline on every producing command: read the bundle, generate, validate, fix, revalidate, confirm zero violations, then output.
Install
From the public marketplace (mirrored to minlang-releases on every release):
/plugin marketplace add codeshift-ai-solutions/minlang-releases
/plugin install minlang@minlang-releases
From a local checkout of this repository:
/plugin marketplace add ./packaging/claude-plugin
/plugin install minlang@minlang-releases
Or load the plugin directory for a single session:
claude --plugin-dir ./packaging/claude-plugin/plugins/minlang
ml1 can print these install commands for any agent without Claude Code:
ml1 language claude-plugin
Command families at a glance
| Family | Commands | Page |
|---|---|---|
| Authoring | author-app, add-feature, model-domain, design-ui, validate-authority | Commands → Authoring |
| Migration | migrate-project, migrate-domain, migrate-screen, migrate-tests, migration-ledger | Commands → Migration |
| Refactoring | pull-domain-logic, dedupe-rules, thin-shell, generated-drift | Commands → Refactoring |
| Diagnostics | explain-failure, why-not-supported, inspect-graph, release-check | Commands → Diagnostics |
| Creative | product-interview, legacy-strangler, boundary-police, demo-from-idea, bundle-upgrade-coach | Commands → Creative |
What every producing command guarantees
Each command that produces or validates MinLang follows the same contract:
- Load the authority-preflight skill first.
- Read the applicable language bundle (stable URL or the project’s pin) before output.
- Detect project shape (
minlang.json, primary.ml,Makefiletargets) andml1availability. - Run validation before declaring success when MinLang is produced or changed.
- Never hand-edit
app/generated/and never write.env. - Never emit the zero-tolerance tokens
now(),today(),random(),current_user. - Return a concise report: changes, verification run, blockers, next step.
Components
| Directory | Purpose |
|---|---|
commands/ | The 23 slash command definitions and their commands.json inventory |
skills/authority-preflight/ | Shared bundle-load + project-detection + verification workflow |
agents/ | Optional creator, gatekeeper, and migration-architect sub-agents |
hooks/ | Non-destructive post-edit .ml validation reminder |
mcp/ | Optional local ml1 MCP bridge (allowlisted read-only commands) |
Authority
Every producing command loads the authority-preflight skill and reads the stable language bundle before output:
https://github.com/codeshift-ai-solutions/minlang-releases/releases/latest/download/minlang-language-bundle.md
The bundle is the single source of truth. If anything in the plugin (or on this site) disagrees with the bundle, the bundle wins.
Source and validation
The plugin lives in this repository under packaging/claude-plugin/. Maintainers validate it with:
make check-claude-plugin
make test-claude-plugin
make verify-claude-plugin-release
Where to go next
- Commands — every slash command, what it does, and how it verifies.
- Agents & tooling — sub-agents, the skill, the hook, and the MCP bridge.
- CLI reference — the
ml1commands these workflows run. - Getting started — install
ml1and scaffold an app.