Anthropic SKILL.md package that teaches an LLM how to author OpenExpertise experiences.

TypeScript types + JSON Schema + parser/validator for the OpenExpertise experience.yaml format.

Dispatcher for `experience` nodes — nests one OpenExpertise experience inside another.

Scheduler, dispatcher registry, event bus, and SQLite-backed state store — the OpenExpertise runtime.

OpenAI-compatible LLMClient implementation for OpenExpertise — wraps any OpenAI-API endpoint (OpenAI, Azure, self-hosted).

`oe-mcp` MCP server — exposes 6 OpenExpertise tools so external MCP-aware agents can run experiences.

Schema-aware authoring helpers — validate, scaffold, and edit experience.yaml files programmatically.

The `oe` command-line interface — run, resume, inspect, evolve, validate experiences.

Advisor that reads a run's events+state and proposes graph upgrades as git-apply-ready diffs.

Ink-based terminal UI for `oe run --tui` — live phases, node status, token stream, activity feed.

Dispatcher for `dataset` nodes — loads file/sqlite/http/MCP-resource data into state.

Dispatcher for `skill` nodes — invokes Anthropic-style SKILL.md packages as graph steps.

Dispatcher for `tool` nodes — runs deterministic JavaScript/TypeScript impl files.

Dispatcher for `agent` nodes — calls an LLM with structured-output schema, writes validated keys to state.

Dispatcher for `cli-agent` nodes — delegates work to Claude Code, Codex, or Gemini CLI subprocesses with timeout/retry.