Maude

A personal marketplace of Claude Code plugins. Two plugins today, plus a maude CLI for scaffolding and running the bundled dev tooling.

📚 Full docs: https://maude.iagh.cz (or browse the source under site/content/docs/ until the public URL lands). Contributing? See CONTRIBUTING.md. Security? See SECURITY.md.

| Plugin | What it does | | ------ | ------------ | | design | Canvas-first iteration on TSX/JSX mocks under .design/ — element selection via Cmd+Click, auto-managed dev server, chained UX/DS critique. | | flow | Generic agentic workflow loop with a second-brain .ai/ workspace. /flow:plan, /flow:execute, /flow:utils-verify, /flow:validate, /flow:done, /flow:init, /flow:record-ddr, /flow:scenario, …. Project-agnostic via <project> placeholders + per-repo .ai/workflows.config.json. |

Plus the maude CLI — maude init scaffolds a fresh .ai/ workspace from the flow plugin skeleton; maude design serve boots the design dev server. The legacy mdcc alias still works (prints a deprecation warning) and will be removed in v0.17.x.

Quick start

1. Add the marketplace inside Claude Code

/plugin marketplace add 1aGh/maude

2. Install the plugins you want

/plugin install design@maude
/plugin install flow@maude

Then /reload-plugins and you should see /design:edit, /design:*, /flow:plan, /flow:execute, etc.

3. Install the CLI

# From npm:
npm i -g @1agh/maude

# Or directly from GitHub:
npm i -g github:1aGh/maude

After install you have these bins on $PATH:

• maude — the primary CLI (init, config, design serve).
• mdcc — legacy alias for maude. Prints a deprecation warning; will be removed in v0.17.x.

4. Bootstrap a repo

In any project root:

maude init                          # scaffold .ai/ second-brain workspace

Then inside Claude Code (with flow@maude installed):

/init                  # Anthropic's built-in — generates CLAUDE.md tailored to your codebase
/flow:init    # populates .ai/workflows.config.json with detected stack
/flow:status           # confirm everything wired up

/init writes the CLAUDE.md Claude auto-loads every session (conventions, build commands, gotchas). /flow:init handles the structured workspace config — they're complementary, not duplicates.

Health check — maude doctor

maude doctor reports missing dependencies, config schema errors, stack drift, and missing quality-gate declarations in one shot. --fix applies safe auto-fixes (prompts per dependency install; never overwrites existing user config values). --json for programmatic consumers. Run it after stack changes or before opening a PR.

maude doctor          # full health check (deps + config + quality)
maude doctor --fix    # install missing deps (per-item prompt) + apply safe config fixes
maude doctor --json   # machine-readable envelope

Quality gates are declared in .ai/workflows.config.json under the top-level quality map — a flat gate → shell command mapping (e.g. { "lint": "pnpm lint", "tests": "pnpm test" }). Flow commands (/flow:validate, /flow:utils-verify, /flow:quick) read it directly and run each via eval. There is no maude quality run wrapper — pnpm <script> is already the runner. Hook it into your own pre-commit tool with a one-liner: eval $(jq -r '.quality.lint' .ai/workflows.config.json). Populate the block by running maude doctor --fix once your package.json scripts exist (additive — it never overwrites a command you tuned).

Sidecar cache — maude cache

Flow + design commands reuse expensive cross-session work through a small sidecar cache at .ai/cache/ (DDR-061): domain research (skips 30–90 s of WebSearch on a same-domain brief), codebase-intelligence scans (skips rescan when the tree is unchanged), parsed design-system vocabulary, and security-review reuse (one shared 1-hour window across /flow:validate, /flow:done, /flow:validate-security). Correctness comes first — most layers are content-addressed (a changed input changes the key → guaranteed miss), and a stale entry is never served speculatively.

maude cache list              # layers, entry counts, sizes, last-write time
maude cache stats             # hit/miss counters + hit-rate per layer
maude cache inspect <layer>   # list a layer's entries; add a <key> to print one
maude cache clear [layer]     # wipe one layer (or everything)

The research/domain and codebase-intelligence layers are committed (deterministic on a given tree, so collaborators share the hit); the per-brief, design-context, security, and scenario layers are gitignored. maude init drops a .ai/.gitignore that encodes the split.

Plugins call maude for executable logic

Plugin slash-commands reach all executable logic through the on-PATH maude binary — never a relative cli/lib/*.mjs path or a raw $CLAUDE_PLUGIN_ROOT/dev-server/bin/*.sh invocation (DDR-062). The marketplace copies each plugin alone (no sibling cli/, no dev-server/), and a flow command's $CLAUDE_PLUGIN_ROOT points at the flow plugin (which has no dev-server at all) — so the only contract that holds across every install shape is maude, which resolves bundled helpers from its own package root. Cache/preflight go via maude cache … / maude preflight …; the design dev-server's bash helpers go via maude design <verb> (screenshot, server-up, prep, slug, smoke, runtime-health, …) — see maude design help. Keep the global maude current; a stale binary means stale helpers.

Runtime requirements

• Node ≥ 20 — for the dev server and CLI. Zero npm runtime deps.
• Claude Code — desktop app, CLI, or IDE extension.
• Optional: agent-browser for design screenshot evidence.

Collaboration model

Two clean paths, no middle ground (DDR-047):

• v1.0 — git handoff OR loopback multi-tab. Push / pull is the cross-machine story. On a single machine, two browser tabs (or two Claude Code instances editing the same repo) sync cursors, comments, and annotations live over loopback WebSocket. The dev server refuses any non-loopback host header on the collab WS endpoint.
• v1.1 — deploy a hub (Phase 9, in-flight). Cross-machine live collab needs a hub binary you deploy yourself. No tunnel mode; no shared cloud.

Security

Solo mode (the default) is fully local — no accounts, no telemetry, no network trust surface, and the canvas sandbox is on by default. Linked (hub) mode is opt-in and carries a documented trust model with disclosed, bounded residuals (hub-pushed content is treated as untrusted input). The full account — what runs where, what the canvas sandbox does and doesn't close, .tsx sync's double opt-in, untrusted-context handling — is at /docs/security (source: site/content/docs/security.mdx). To report a vulnerability, see SECURITY.md.

What's where

User-facing docs live in two places — the README points you the right way:

• Reference (every command, every config key, recipes for Next.js / Expo / monorepo) → site/content/docs/ (served at https://maude.iagh.cz once Vercel is wired — see DDR-005).
• Quickstart + contributor info → this README.

The docs site auto-generates per-command pages from plugins/{flow,design}/commands/*.md frontmatter and a typed schema reference from plugins/flow/.claude-plugin/config.schema.json. Adding a new command → docs update on next build.

Workspaces

The repo is a pnpm workspace monorepo with one published npm package (@1agh/maude). Internal workspaces are "private": true and never publish:

| Workspace | Purpose | | --------- | ------- | | . (root) | The single npm publisher — CLI, dev-server entry, plugin templates that ship to npm. | | site/ | Docs site — Fumadocs + Next.js, deployed to Vercel (DDR-005). | | apps/studio/ | Zero-dep Node dev server + browser client. Bundled output (dist/) is the only thing in the npm tarball. | | apps/hub/ | Reserved for the v1.1 federated hub (Phase 9). |

Common scripts at the root:

pnpm install          # bootstrap everything
pnpm dev              # boot the design dev server
pnpm dev:site         # docs site dev server
pnpm build            # build every workspace that defines `build`
pnpm lint             # biome over the whole repo
pnpm test             # node --test over cli/**/*.test.mjs
pnpm changeset        # add a changeset for the next release

The site workspace's Next.js dependencies are heavy — contributors fixing plugin code can keep working with pnpm install --filter '!@maude/site' to skip them.

Updating

In Claude Code, after a new version lands on the marketplace:

/plugin marketplace update maude
/plugin install design@maude
/plugin install flow@maude

Releasing

The npm package (@1agh/maude) and the Claude Code plugins (design@maude, flow@maude) share one version. The standard release path is Changesets:

# 1. Each PR with shipped behavior includes a changeset
pnpm changeset

# 2. When ready to release, the maintainer runs:
pnpm version            # = bash scripts/changesets-version.sh
                        #   → consumes .changeset/*.md
                        #   → bumps package.json + CHANGELOG
                        #   → propagates version to plugin manifests
                        #   → re-runs scripts/check-version-parity.sh

git commit -am "chore: release v$(node -p "require('./package.json').version")"
git tag "v$(node -p "require('./package.json').version")"
git push --follow-tags

The v* tag triggers .github/workflows/publish.yml, which re-runs the parity check, verifies the tag matches package.json, builds workspaces, publishes with --access public --provenance, and creates a GitHub Release using the CHANGELOG entry for that version.

scripts/bump-version.sh patch|minor|major|X.Y.Z remains as a manual fallback for emergency hotfixes outside the Changesets flow.

One-time setup (project owner)

1. Create an Automation token at https://www.npmjs.com/settings/~/tokens.
2. GitHub repo → Settings → Secrets → Actions → NPM_TOKEN.
3. id-token: write is already enabled in publish.yml for npm provenance.

Local development (plugin authors)

/plugin marketplace add /absolute/path/to/maude
/plugin install design@maude
/plugin install flow@maude

Working on plugin internals:

1. Edit in place — the local marketplace points at your working tree.
2. Reload after edits:
   • Commands / agents / skills → /plugin marketplace update maude then /reload-plugins.
   • Dev server code → kill the running process (lsof -i :<port> → kill) and let the next /design:edit invocation auto-restart.
3. Test in isolation — open Claude Code from a scratch project (cd /tmp && claude) so plugins aren't entangled with this repo's own .ai/.
4. Dogfood — Maude itself uses flow for plan/execute/done. Once flow is installed against this marketplace, you can drive its own development with /flow:plan, /flow:execute, etc.

License

MIT — see LICENSE.