Release prep: v6.4.0 notes and version bump

[obra]

This PR targets dev.

Who is submitting this PR? (required)

Field	Value
Your model + version	Claude Fable 5.1 (claude-fable-5-1)
Harness + version	Claude Code 2.1.276 (CLI, macOS)
All plugins installed	Installed per claude plugin list (not all enabled): agent-sdk-dev@claude-plugins-official, clangd-lsp@claude-plugins-official, claude-code-setup@claude-plugins-official, claude-opus-4-5-migration@claude-code-plugins, claude-pa@claude-pa-dev, claude-session-driver@superpowers-marketplace, code-review@claude-plugins-official, code-simplifier@claude-plugins-official, context7@claude-plugins-official, cxdb-observability@local, document-skills@anthropic-agent-skills, double-shot-latte@superpowers-marketplace, elements-of-style@superpowers-marketplace, episodic-memory@superpowers-marketplace, example-skills@anthropic-agent-skills, feature-dev@claude-code-plugins, feature-dev@claude-plugins-official, frontend-design@claude-code-plugins, frontend-design@claude-plugins-official, github-triage@github-triage-dev, gopls-lsp@claude-plugins-official, ledger-memory@ledger-memory-market, linear@claude-plugins-official, mcp-server-dev@claude-plugins-official, plugin-dev@claude-code-plugins, plugin-dev@claude-plugins-official, pr-review-toolkit@claude-plugins-official, primeradiant-ops@primeradiant, release-radar@whats-new-marketplace, rust-analyzer-lsp@claude-plugins-official, security-guidance@claude-plugins-official, summarize-meetings@2389-research-marketplace, superpowers-chrome@superpowers-chrome-dev, superpowers-chrome@superpowers-marketplace, superpowers-dev@superpowers-marketplace, superpowers-developing-for-claude-code@superpowers-marketplace, superpowers-lab@superpowers-marketplace, superpowers@claude-plugins-official (6.3.0), swift-lsp@claude-plugins-official, worldview-synthesis@2389-research-marketplace
Human partner who reviewed this diff	Jesse Vincent (@obra) asked for the draft and for this PR, and reviews the diff here. He had seen a summary of the notes, not the complete diff, when this was opened, which is why it is a draft.

What problem are you trying to solve?

19 PRs have landed on dev since the v6.3.0 merge-back (777ceb5), including two new skills and a rebuilt executing-plans, and RELEASE-NOTES.md has no v6.4.0 section. The release flow's first step is a release-prep PR against dev with the notes. Jesse asked for the notes to be drafted in a worktree off dev.

What does this PR change?

Adds a ## v6.4.0 section to the top of RELEASE-NOTES.md, in the v6.3.0 style, and bumps every declared manifest to 6.4.0 with scripts/bump-version.sh 6.4.0 (audit clean). The OpenCode pin examples in .opencode/INSTALL.md and docs/README.opencode.md now point at v6.4.0, the first release with V2 support.

Is this change appropriate for the core library?

Yes. It is the project's own release notes.

What alternatives did you consider?

• Generating the notes from commit subjects. Rejected: the subjects don't say what a user will notice, and several entries (inline execution, Review Focus, the exec-bit fix) only make sense with the PR's problem statement. I read each PR's body and the resulting skill text instead.
• Bundling the version bump into this PR, as the release flow allows. Left out because Jesse asked for a draft of the notes and the release date isn't settled.

Does this PR contain multiple unrelated changes?

No. One file, one new section.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs: none found (searched open and closed for "release notes 6.4" and "v6.4.0"). The section summarizes feat(opencode): support OpenCode 2.0.4+ alongside V1 #2106, feat(tdd): the project's suite defines green, not just your test file #2110, Update to Prime Radiant Community Code of Conduct. #2122, docs: add Qwen Code install instructions to README #2132, fix(requesting-code-review): anchor the multi-commit BASE_SHA alternative to the merge base #2133, fix(sdd): invoke sdd-workspace via bash so helpers survive stripped exec bits #2134, docs(testing): describe the Quorum eval lab accurately, replacing stale Drill references #2135, fix(sdd): reject empty or non-descendant BASE..HEAD ranges in review-package #2136, fix(sdd): ownership markers stop same-basename plans sharing a workspace #2138, feat(skills): import proving-it-works-with-a-movie skill #2214 (with feat(movie): support native Windows PowerShell and Git Bash #2275), feat: add diagnosing-superpowers skill — evidence-based session diagnosis, scrubbed bug-report bundles #2236, Fix platform-support issue template to apply a label that exists #2250, fix: preserve shared intent through design and planning #2258, fix: preserve diagnostic evidence in doctor exports #2287, Invoke bundled scripts through their interpreter in skill prose #2301, feat: add native Muse support (multiprovider) #2317, Rebuild executing-plans as a first-class inline (Native) execution mode #2318, Reviewer judges the spec as a vision document; plans list the five implied cases most likely to bite #2319, Add Claude Code platform reference: a nested orchestrator for cheaper subagent-driven development #2320.

Environment tested

Harness (e.g. Claude Code, Cursor)	Harness version	Model	Model version/ID
Claude Code	2.1.276	Claude Fable 5.1	claude-fable-5-1

Nothing executable changed, so "tested" here means the claims were checked against the repo: the PR list comes from git log --first-parent 777ceb5..origin/dev, and the statements about skill behavior were checked against the skill files on dev (for example, the writing-plans handoff wording, the TDD paragraph, CLAUDE.md now pointing at AGENTS.md).

New harness support (required if this PR adds a new harness)

Not applicable. This PR adds no harness; it describes harness PRs that already merged with their own transcripts.

Evaluation

• Initial prompt: "in a new worktree off dev, could you draft the release notes for superpowers 6.4, please?"
• Eval sessions after the change: none. This is documentation with no behavior-shaping content.
• Outcome change: not applicable.

Things the reviewer should check

• Date. The heading says 2026-09-17 as a placeholder for the real release date.
• "The old batch-with-human-checkpoints behavior is gone." Stated outright under Executing Plans because users who relied on it should hear it. Reword if that overstates it.
• Numbers. "1 of 12 probe runs" (feat(tdd): the project's suite defines green, not just your test file #2110) and "about half the cost and wall clock" (Add Claude Code platform reference: a nested orchestrator for cheaper subagent-driven development #2320) are taken from those PRs' bodies. No other figures are cited.
• Movie skill dependencies. "Needs uv and ffmpeg" comes from counting mentions in the skill's docs, not from a declared requirements list. Piper and Playwright appear for some routes and are not mentioned. @arittr would know whether that is the right summary.
• Left out as internal: docs/superpowers specs and plans, and the inline-eval tooling.

Rigor

• If this is a skills change: I used superpowers:writing-skills and completed adversarial pressure testing (not a skills change)
• This change was tested adversarially, not just on the happy path (not applicable to release notes)
• I did not modify carefully-tuned content (Red Flags table, rationalizations, "human partner" language) without extensive evals showing the change is an improvement

Human review

• A human has reviewed the COMPLETE proposed diff before submission

Unchecked on purpose: Jesse directed the PR and reviews the diff here. It stays a draft until he has.

[arittr]

LGTM, add note about AGENTS vs CLAUDE