Skip to content

MDA Open Spec v1.0.0-rc.4 — five runtimes verified

Latest
Latest

Choose a tag to compare

View all tags
@LarHope LarHope released this 26 May 04:27
v1.0.0-rc.4
cbd6ead

What's new

Five runtimes verified end-to-end. An MDA-emitted SKILL.md now loads cleanly in Claude Code, Codex CLI, OpenCode, Hermes Agent, and OpenClaw with reproducible install kits under compat/. Each kit ships a verify.sh that re-compiles the source and validates against the JSON Schema, an install.sh that lands the artifact in the runtime's own discovery path, and a notes.md recording the empirical proof plus any per-runtime authoring gaps.

Spec §06-3.4 scopes the v1.0 SKILL.md envelope. v1.0 now explicitly declares that runtime-specific top-level frontmatter fields (the eleven Claude Code extensions, OpenClaw's user-invocable, similar one-off keys from other runtimes) sit outside the v1.0 schema. They graduate to v1.1 once the schema expansion lands. This unblocks Tier B compatibility claims across runtimes without watering down what the v1.0 envelope guarantees.

README rebuilt for the audience, not the contributor. The Compatibility section moved above the fold, with a hand-crafted SVG card grid (one card per runtime, brand-color accent, real product logo, "drop-in compatible" eyebrow, mobile-wrapping layout). Engineer-prose ecosystem paragraphs trimmed back to the AGENTS.md reach claim in a single sentence.

Other changes

  • apps/cli shipped as @markdown-ai/cli@1.1.7 via npm Trusted Publishing (OIDC, sigstore-signed provenance, no NPM_TOKEN). Adds mda demo for zero-friction first run, --version, native LLMix registry root support.
  • CI now auto-creates a GitHub Release on every successful CLI publish.
  • CLAUDE.md codifies the CLI release runbook and mandates hand-written GitHub Release notes per version.

What's not in this release

This remains a v1.0 release candidate. The conformance suite, the reference verifier, the cross-runtime resolver, and the v1.1 schema expansion for runtime-specific top-level fields stay pending. The full punch list lives in docs/v1.0/what-v1.0-does-not-ship.md.

Try it

npm install -g @markdown-ai/cli
mda demo
mda compile your-source.mda --target SKILL.md --integrity

To verify the compatibility claim yourself, clone and run any of the per-runtime kits:

git clone https://github.com/sno-ai/mda
./compat/claude-code/verify.sh
./compat/codex-cli/verify.sh
./compat/opencode/verify.sh
./compat/hermes/verify.sh
./compat/openclaw/verify.sh
Assets 2
Loading