v0.1.4 — ultra prompt hardening + DX sweep + OE_OPENAI_MODEL override

0.1.4 — 2026-05-28

Patch release bundling 4 user-facing improvements: 2 DX fixes from the v0.1.3 post-release sweep, the OE_OPENAI_MODEL / OE_ANTHROPIC_MODEL override unblocking self-hosted users, and a substantial synthesizer-prompt hardening for oe ultra. All caught from real-API testing.

Added

• OE_OPENAI_MODEL and OE_ANTHROPIC_MODEL env vars — defaultModelFor now honors these as overrides. Previously oe ultra and oe evolve hardcoded gpt-4o-2024-11-20 and threw 404 against any self-hosted OpenAI-compatible endpoint (vLLM / Ollama / LM Studio / etc.). Per-node YAML model: overrides still take precedence; the env vars apply when no per-node model is set.

Changed

• oe ultra synthesizer prompt — substantial hardening. Live-tested against a self-hosted MiniMax-2.7-w8a8 endpoint; the first synthesis attempt produced YAML with 3+ validation errors. Six iterations of prompt improvements landed:

  1. Top-level shape rule — nodes: and edges: MUST nest under graph:, not at root
  2. YAML escaping rule — always quote strings containing :, #, {}, [], etc.
  3. Per-kind required-field cheatsheet (one block per node kind, dataset/cli-agent fully spelled out)
  4. Edges have ONLY from/to/when — no description:, no label:
  5. state.schema type values restricted to string | number | boolean | object | array | null (no integer, no int)
  6. Dataset semantics — kind: dataset is for tabular rows (JSON/JSONL/CSV/Parquet/SQLite) only; for single-file loads use kind: tool with readFileSync; writes: MUST be exactly one field
  7. Defensive tool stubs — oe run MUST succeed before any user wiring; every state read uses ?? fallback; the first node populates everything downstream needs from a fixture
  8. Explicit 11-item self-check before returning

  Reduced failure rate from "100% need manual YAML fixes on MiniMax-2.7-w8a8" to "schema-clean and runtime-mostly-clean on first try". The remaining hard case is cross-field consistency (writes: matching state.schema names) — best solved by auto-retry on validation failure, planned for the next release.

• The clone destination for subpath-based gh: installs uses the last meaningful path segment as the name, so gh:org/monorepo/examples/x and gh:org/monorepo/examples/y co-exist in .openexpertise/experiences/x/ and …/y/ instead of colliding.

Fixed

• oe validate <non-existent-path> now prints a structured experience.yaml not found at <path> message instead of a bare ENOENT propagating from the file-system read. Matches the error style of every other command. (Reported by the v0.1.3 patch sweep.)
• oe install gh:owner/repo/subpath/path now parses the trailing path components as a subpath instead of silently dropping them. gh:jane/monorepo/examples/digest@v0.2.0 clones jane/monorepo at v0.2.0 and installs only examples/digest/ into .openexpertise/experiences/digest/. (Sweep finding #2.)

v0.1.3 — oe ultra polish + 3 cookbook recipes + 2 tutorials + 5 case studies

0.1.3 — 2026-05-28

Same-day patch — UX polish on the unique features (oe ultra) + cookbook + tutorial work that landed after 0.1.2. No breaking changes.

Added

• oe ultra two-phase progress indicator — replaces the silent 30-90s wait with per-phase spinner + ✓ + timing, so users know whether the LLM is alive between analyze and synthesize.
• oe ultra --dry-run flag — runs Phase 1 (analyze) only, prints the detected shape (nodes, phases, open questions), exits 0 without spawning the heavier Phase 2 synthesis. Halves the time + tokens for iterating the task description.
• Numbered next-steps output on oe ultra completion — replaces the pino JSON dump with a human-readable checklist + reference link to the docs tutorial.
• UltraExpertise.author() opts gain optional onPhase callback (typed PhaseEvent union) and stopAfterAnalyze flag. Backwards-compatible — passing neither preserves prior behavior.
• 3 new cookbook recipes for v0.1.1 capabilities:
  • mcp-resource-dataset — reading from an MCP server via the dataset source. Covers mcp.json config, row normalization contract, and the 5 actionable error paths.
  • submit-to-registry — happy path for oe submit with the pre-submit checklist + field auto-detection table + dry-run.
  • cross-vendor-chain — Claude → Codex → Gemini serial pattern extracted as a reusable recipe.

Docs

• "Your first experience" tutorial at /guide/first-experience — full 30-min walkthrough from oe init through tested + registry-submittable flow. Companion examples/your-first-experience/ reference implementation (weekly engineering digest).
• "Authoring with oe ultra" tutorial at /guide/authoring-ultra — parallel 30-min walkthrough for the ultra-driven authoring path. Different domain (PR welcome bot) to avoid overlap.
• 5 case-study deep dives at /use-cases/:
  • Multi-dimensional PR review (engineering)
  • Executable on-call runbook (SRE)
  • Multi-vendor compliance scan (security)
  • Reproducible LLM evaluation suite (ML platform)
  • Tier-1 support routing (customer ops)

Changed

• README install section adds a callout for oe ultra — pointing first-time users at the LLM-authored path.
• Site nav version dropdown bumped 0.1.0 → 0.1.2 → 0.1.3.
• Test count: 293 → 310 (+17 in packages/cli/tests/ultra.test.ts).

v0.1.2 — fix oe --version

0.1.2 — 2026-05-28

Tiny patch over 0.1.1 — the kind of bug only noticed after a fresh npm install.

Fixed

• oe --version now prints the actual installed version. Was hardcoded to 0.1.0 in packages/cli/src/index.ts:23; rewired to read from the bundled package.json at runtime so it never drifts again.

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.

v0.1.0 — First public release

Changelog

All notable changes to OpenExpertise will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.

0.1.0 — 2026-05-27

First public release. OpenExpertise is an AI-era Makefile: codify expert
workflows as YAML graphs, run them with deterministic flow + LLM-powered
nodes, and evolve the graph after each run.

Added

Runtime

• 6 node kinds in a single graph schema: tool (deterministic code),
  agent (LLM with structured output), skill (Anthropic SKILL.md packages),
  dataset (file / SQLite / HTTP / MCP-resource), experience (nested OE),
  and cli-agent (delegate to Claude Code / Codex / Gemini subprocesses).
• Sequential and parallel schedulers with bounded concurrency
  (--concurrency N flag and runtime.concurrency in YAML); topological
  wave execution; 429-aware exponential backoff retry.
• Persistent SQLite state store (.openexpertise/state.sqlite) — every
  node's writes land in a typed blackboard; resume across sessions with
  oe resume <run-id>.
• JSONL event log (.openexpertise/runs/<id>.jsonl) — every dispatch,
  retry, write, and error captured for replay and audit.
• Per-node memoization cache for cheap re-runs after edits.
• for_each fan-out with concurrency honored, plus when: conditional
  edges for branching.
• State merge strategies: array_append, set_once, last_wins.

CLI (oe)

• oe run <experience> with --tui, --concurrency, --resume, --once.
• oe inspect <run-id> — event-ordered run reconstruction (parallel-safe sort by ts).
• oe state <field> — pull any field out of state SQLite.
• oe resume <run-id> — replay from the last successful node.
• oe validate <experience> — schema check before running.
• oe evolve <run-id> — advisor writes proposal markdown with git-apply-ready diff.
• oe ultra "<intent>" — one-keyword authoring: LLM scaffolds a full experience from a sentence.

CLI agent integration

• Subprocess runner with timeout, retry, output-format parsing (text | json),
  and AJV schema validation against parsed JSON.
• Supported providers: claude-code, codex, gemini.
• Two-way: outbound (delegate node to a CLI agent) AND inbound via oe-mcp
  (5 MCP tools exposed so external agents can run experiences from their sessions).

Authoring

• Schema-aware authoring helpers in @openexpertise/authoring.
• /ultraexpertise slash command + matching oe ultra CLI.
• Anthropic SKILL.md package (@openexpertise/skill-experience-creator) that
  teaches a code-assistant LLM how to author OE experiences.

TUI

• Ink-based live dashboard: phase progress, per-node status, live token stream,
  activity feed of recent events. Toggle with --tui.

Built-in examples (11)

• hello-tool — smallest possible flow.
• dataset-aggregate — CSV → aggregate.
• agent-echo — single agent with structured output.
• review-branch ★ — multi-dim code review + verifier + score + evolution. The hero demo.
• oncall-runbook — incident triage via for_each fan-out.
• issue-triage — classify → search dupes → conditional dedup → route. Shows when: edges.
• release-gates — license + changelog + coverage + Claude-Code security scan → release gate.
• cli-orchestration — Claude Code summarizes; Codex critiques.
• tri-cli-orchestration ★ — Claude → Codex → Gemini in one DAG.
• deep-research — Claude Code WebSearch + Gemini Google Search → cited synthesis.
• systematic-debugging — translates the superpowers systematic-debugging skill into a YAML flow.

Tests

• 227 passing across 58 test files. Every example ships a mocked-LLM e2e test.

Docs

• README with 60-second demo, comparison vs LangGraph/CrewAI/Anthropic workflows/Claude Code.
• Per-example README with run instructions and ASCII pipeline diagram.
• docs/comparison.md deep-dive vs alternatives.
• docs/superpowers/ design diaries (one per major plan: 1-6 and A-F).
• CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md.

Acknowledgements

• The systematic-debugging example is a direct translation of the
  Anthropic superpowers skill of the
  same name — reused with attribution.
• The TUI uses Ink by Vadim Demedes.