runx-runtime 0.6.19

Native Rust runtime for local runx execution, adapters, harness replay, receipts, and sandboxing.
Documentation

runx-runtime

Native Rust runtime for governed runx execution.

This crate owns the canonical local orchestration path for Rust-backed runx: skill execution, graph execution, harness replay, host reporting, sandbox preparation, receipts, history projection, adapters, and domain-free effect orchestration. Pure parser, core, contract, receipt, and domain crates remain upstream.

Current slice:

  • parses a local graph with runx-parser
  • plans sequential/fanout transitions with runx-core
  • runs cli-tool skills behind the cli-tool feature
  • emits receipts and validates the parent receipt tree with runx-receipts
  • exposes native skill, doctor, list, history, MCP, registry, config, policy, tool, and dev command support through runx-cli

Adapter families remain feature gated:

  • cli-tool
  • mcp
  • mcp-http-server
  • a2a
  • agent
  • catalog
  • external-adapter
  • http

a2a is contract-defined but not enabled in runx-cli; the CLI enables cli-tool, catalog, mcp, mcp-http-server, external-adapter, agent, and http.

Doctor

The native Rust doctor API is wired into runx-cli for the read-only diagnostic surface. It must not shell out to npm or TypeScript for canonical local behavior.

This crate currently ports the read-only fixture-backed diagnostics:

  • runx.tool.manifest.removed_format
  • runx.tool.fixture.missing
  • runx.skill.fixture.missing
  • runx.structure.file_budget.exceeded
  • runx.structure.cross_package_reach_in

Deferred doctor families remain owned by follow-up slices:

  • runx doctor --fix repair writes
  • diagnostic catalog, --list-diagnostics, and --explain
  • official skills lock freshness
  • tool manifest stale source and schema hashes
  • packet index diagnostics
  • graph packet path validation
  • receipt proof health
  • policy health