Skip to content

feat(cua-driver): implement SDK-owned runtime RFC - #2561

Merged
f-trycua merged 21 commits into
mainfrom
codex/fix-e2e-unrestricted-runtime
Jul 25, 2026
Merged

feat(cua-driver): implement SDK-owned runtime RFC#2561
f-trycua merged 21 commits into
mainfrom
codex/fix-e2e-unrestricted-runtime

Conversation

@f-trycua

@f-trycua f-trycua commented Jul 24, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • implement RFC 2549’s SDK-owned runtime architecture while preserving CLI/MCP as the preferred standalone-agent integrations
  • add immutable runtime/session authorization, direct and private-worker ownership, remote envelope negotiation/cancellation, authenticated service sessions, and deterministic shutdown semantics
  • preserve standalone macOS TCC ownership while adding explicit cua-driver mcp --direct for hosts that deliberately own attribution
  • harden local service and Windows UIAccess boundaries, fail closed in Windows Session 0, and make platform-facility limitations structured and explicit
  • retain the earlier browser-scale and nested-Wayland E2E fixes already carried by this PR

Compatibility

  • released CLI/MCP defaults, constructors, package exports, schemas, and result envelopes remain covered by frozen fixtures
  • previous-release Rust, Python, and TypeScript application sources compile or run unchanged against the candidate packages
  • daemon-backed CLI/SDK actions negotiate compatibility before dispatch, including replacement daemons at the same endpoint
  • legacy remote-carrier implementations remain source-compatible but fail closed until they advertise cancellation support

Documentation and examples

  • document when imported applications should use direct SDK ownership versus standalone CLI/MCP
  • document macOS TCC attribution, Windows interactive-session requirements, Linux display ownership, service authentication, and runtime lifecycle
  • update the reusable Cua Driver skill and in-process SDK examples
  • add previous-release application fixtures for Rust, Python, and TypeScript

Local and standard CI verification

  • cargo check --workspace --all-targets --locked
  • SDK, CLI compatibility, private-worker, macOS overlay, contract, core parity, and browser setup-UI suites
  • generated contract, UniFFI binding, and public C ABI checks
  • TypeScript typecheck and SDK tests
  • Python UniFFI loader tests
  • agent SDK Python/TypeScript example checks
  • docs internal/external links, public hygiene, and production build
  • release metadata, platform unit/compile, Nix policy/package, script, attribution, and fleet-mirror checks
  • git diff --check and sensitive-content scan

Independent review

Claude Code Opus 5 reviewed the complete RFC diff and follow-up macOS setup-UI hardening. All blocking findings were fixed; the final verdict is GO.

Exact-SHA certification

Candidate: 9cf8db8c73966529f32e2ac7f574181f5460e188

This PR is ready for review and intentionally left unmerged.

@f-trycua f-trycua changed the title fix(cua-driver): complete browser and nested Wayland E2E feat(cua-driver): implement SDK-owned runtime RFC Jul 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant