v0.1.1 — oe demo + init templates + oe submit + mcp-resource wired

0.1.1 — 2026-05-28

First patch release. Closes the most-visible Day-1 gaps: a zero-token onboarding demo, real starter templates, frictionless registry submission, and the mcp-resource dispatcher that the schema already accepted.

Added

• oe demo — preview a bundled example without an API key. Ships 4 pre-recorded runs: deep-research, review-branch (★ with built-in advisor evolution proposal), systematic-debugging, brainstorming. The ★ one demonstrates the author → run → evolve loop in 60 seconds.
• oe init --template <name> — scaffold from one of 4 real starter shapes:
  • tool-only (default; preserves prior behavior — single tool node, no LLM)
  • agent — single LLM agent with structured output
  • cli-agent — single Claude Code subprocess invocation
  • full-pipeline — tool → agent → cli-agent → tool, with state passing between every kind and a Markdown report at the end
• oe submit — zero-friction registry submission. Inside an experience directory, validates the YAML, auto-detects GitHub remote / ref / subpath, generates the canonical registry.json entry, and opens a pre-filled issue against xingchengxu/OpenExpertise using the experience_submission.md template. Flags: --tags, --name, --ref, --subpath, --description, --dry-run, --output <file>.
• mcp-resource dataset source now actually works. Spawns an MCP server (stdio) per mcp.json config and reads resources via resources/read. Previously the schema accepted the shape but the dispatcher threw at runtime. Requires @modelcontextprotocol/sdk (new transitive dependency on @openexpertise/node-kinds-dataset). Stdio transport only in 0.1.x; HTTP transport planned for 0.2.x.

Changed

• Test count: 265 → 292 (+27 across the four new commands).
• Pre-publish smoke script (scripts/pre-publish-smoke.sh) now runs all 9 gates in 2-3 minutes end-to-end; required for v0.1.1 republish.

Docs

• Ecosystem positioning page at /ecosystem — explains where OE fits relative to MCP / Skills / Anthropic /workflows / autonomous agents (Claude Code, Codex, Gemini CLI, OpenHands, AutoGen, CrewAI, OpenClaw, Hermes Agent, OpenHuman, …). Honest about the cli-agent provider roadmap.
• /cookbook — 10 self-contained recipes for common patterns (fan-out, retry, conditional edges, merge strategies, nested experiences, etc.).
• awesome-openexpertise.md — curated community list at repo root.
• docs/registry.md — new ## Submit via oe submit section explaining the recommended contribution path.
• README hero — animated docs/assets/demo.svg (oe doctor → registry → init → run) replaces the static doctor frame.

Fixed

• CI workflow pnpm/action-setup version conflict with root packageManager field — silent failure on every push since at least 17bb137. Same fix as docs.yml.
• Documentation consistency drift: "Five MCP tools" → "6" in 7 places; "14 packages" → "15" in CHANGELOG/CONTRIBUTING; stale test counts; outdated node packages/cli/dist/bin.js invocations in user-facing examples replaced with oe.