Release v6.4.0: diagnosing-superpowers and movie skills, Native execution, OpenCode 2.0, Muse, and Qwen Code support

[obra]

Release train for v6.4.0: merges everything on dev since the v6.3.0 merge-back (777ceb5) into main. Tagging, the GitHub release, and merging main back into dev follow after this merge.

The head is release/v6.4.0: one commit on top of main whose tree is identical to dev at 1c63ca9. Opening the PR straight from dev would list 240 commits, 123 of them already released in v6.3.0's squash. The file diff is the same either way.

Who is submitting this PR?

Field	Value
Your model + version	Claude Opus 5 (claude-opus-5)
Harness + version	Claude Code 2.1.277 (CLI, Linux)
Human partner who reviewed this diff	Jesse Vincent (@obra), maintainer — directed this release

What's in the release

• New skills: diagnosing-superpowers (feat: add diagnosing-superpowers skill — evidence-based session diagnosis, scrubbed bug-report bundles #2236, fix: preserve diagnostic evidence in doctor exports #2287), proving-it-works-with-a-movie (feat(skills): import proving-it-works-with-a-movie skill #2214, feat(movie): support native Windows PowerShell and Git Bash #2275)
• Executing plans: rebuilt as Native (inline) execution; the batch-with-checkpoints behavior is gone (Rebuild executing-plans as a first-class inline (Native) execution mode #2318)
• Writing plans: review the saved plan before execution (fix: preserve shared intent through design and planning #2258), Subagent-driven vs Native handoff (fix: preserve shared intent through design and planning #2258, Rebuild executing-plans as a first-class inline (Native) execution mode #2318), Review Focus section (Reviewer judges the spec as a vision document; plans list the five implied cases most likely to bite #2319)
• Brainstorming: intent before features, approval tied to design and planning (fix: preserve shared intent through design and planning #2258)
• Code review: reasonable-user standard for unspecified behavior, Declined-to-judge list (Reviewer judges the spec as a vision document; plans list the five implied cases most likely to bite #2319); merge-base BASE_SHA (fix(requesting-code-review): anchor the multi-commit BASE_SHA alternative to the merge base #2133)
• TDD: the project suite defines green (feat(tdd): the project's suite defines green, not just your test file #2110)
• SDD: per-plan workspace ownership (fix(sdd): ownership markers stop same-basename plans sharing a workspace #2138), review-package range guards (fix(sdd): reject empty or non-descendant BASE..HEAD ranges in review-package #2136), opt-in nested controller on Claude Code (Add Claude Code platform reference: a nested orchestrator for cheaper subagent-driven development #2320)
• New harness support: OpenCode 2.0 (feat(opencode): support OpenCode 2.0.4+ alongside V1 #2106), Muse (feat: add native Muse support (multiprovider) #2317), Qwen Code (docs: add Qwen Code install instructions to README #2132)
• Fixes: scripts invoked through their interpreter so stripped exec bits don't break skills (Invoke bundled scripts through their interpreter in skill prose #2301, fix(sdd): invoke sdd-workspace via bash so helpers survive stripped exec bits #2134); platform-support template label (Fix platform-support issue template to apply a label that exists #2250); movie terminal test expectation (test(movie): expect 8 frames from the slow-capture film test #2329)
• Docs: Quorum testing docs (docs(testing): describe the Quorum eval lab accurately, replacing stale Drill references #2135), README "When Something Goes Wrong", AGENTS.md canonical, Code of Conduct (Update to Prime Radiant Community Code of Conduct. #2122)

Full notes: RELEASE-NOTES.md v6.4.0 section (#2327).

Verification

• scripts/bump-version.sh --check clean across all 11 declared manifests at 6.4.0 (chore: bump version to 6.4.0 #2330)
• Deterministic suites pass on the release tree: antigravity, brainstorm-server, executing-plans scripts, sdd-workspace, worktree-path-policy, codex (marketplace + package), codex-plugin-sync, devin, diagnosing-superpowers, hermes, hooks, kimi, opencode, pi, proving-it-works (146 tests), shell-lint, systematic-debugging, version-bump, writing-skills
• Known, not blocking: tests/brainstorm-server/server.test.js is flaky (1 in 6 locally) because the server prints its startup JSON before writing state/server-info; scripts/package-codex-plugin.sh refuses linked git worktrees. Both to be fixed after the release.
• Live-LLM batteries (tests/claude-code, explicit-skill-requests) not run for this release

https://claude.ai/code/session_01PRUnVUm4g4EcP9eNjAiT2B

[arittr]

removing the proving with video code