DocsMIT · agent-agnostic

Advanced

Optional depth

Starter template, strict scoring, MCP hints, learning loop, and npm publish — not required for the default workflow.

Starter template

starters/dzine-starter/ is a minimal directory for generating packs beside your app. From the repo root:

Full smoke test
./scripts/first-run.sh

See starters/dzine-starter/README.md for npm scripts (dzine:pack, dzine:strict).

Score modes

ModeFlagChecks
Structural v0(default)HTTP fetch: viewport, main, h1, proof/artifact tokens, primary-action words
Strict--strictStructural + dzine_browser_qa.py preflight (Playwright when installed)
Strict score
npx dzine-skill score --contract build-pack/contract/design-contract.json --url http://localhost:3003 --strict --out build-pack/score-report.json

Structural pass does not prove visual quality — only inspectability and rough clause alignment.

MCP (optional)

MCP tools list routes and return CLI hints for render/score/repair rather than executing them in-process. Prefer committed build-pack/ files for CI and team review.

MCP documentation

DZINE Brain / learning loop

The corpus includes learning modules for run capture and vault organization. This is experimental — use dzine_self_improve.py for reviewed patch proposals, not automatic skill rewrites.

Self-improvement report
python3 skills/dzine/scripts/dzine_self_improve.py --root .

Local API and agent sweep

npx dzine-skill api and npx dzine-skill agent exist for local experimentation. They are not part of the public five-minute path.

Publish CLI to npm

Maintainers with npm publish rights:

Local publish
npm login && ./scripts/publish-npm.sh

Or add NPM_TOKEN to GitHub secrets and run the Publish dzine-skill workflow.

Standalone starter export
./scripts/export-starter-standalone.sh