release(0.2.0-beta.1): cut the first gate-certified release from main

[lidge-jun]

v0.1.0 (2026-07-06) describes 25 skills, 12 hooks, 801 tests. main today ships 28 skills, 21 hooks, 1,631 tests, plus the PR #1 runtime hardening merged as dac77cc7. Users cannot tell what they installed.

Version: 0.2.0-beta.1. Minor bump because main diverged materially; -beta because the packed-install lane is brand new and the MLB 1.0 receipts are deferred. Shipping it as stable 0.2.0 would repeat exactly the failure this work is fixing — publishing a claim the evidence does not carry.

Sequence: version bump across package/manifest/components → regenerate inventory → CHANGELOG section → promote dev to main via PR → exact-head CI + packed-install on the merge commit → dispatch release.yml → verify assets.

Rollback policy: a bad release is superseded by 0.2.0-beta.2, never by deleting the tag — deleting a published tag breaks any installer pinned to --ref.

Plan: devlog/_plan/260815_release_train_production/050_publish_and_close.md.

[lidge-jun]

Closed. v0.2.0-beta.1 is published — the first release cut by the train rather than by hand.

release verify: READY — 0.2.0-beta.1 @ 27ed3b2522a195e2503752edb291d8885c8c35e8
                (deferred receipts allowed and recorded)
published v0.2.0-beta.1 with 3 assets

Release run: 31870901990. Assets: codexclaw-payload-0.2.0-beta.1.tar.gz, SHA256SUMS, candidate-0.2.0-beta.1.json.

Binding verified after publication:

tag_commit = 27ed3b2522a195e2503752edb291d8885c8c35e8
manifest   = 27ed3b2522a195e2503752edb291d8885c8c35e8   OK manifest==tag
main       = 27ed3b2522a195e2503752edb291d8885c8c35e8   OK manifest==main

Manifest contents: six train receipts present, nine MLB 1.0 receipts deferred with reasons, testSuite {pass: 1659, fail: 0, total: 1660} measured on the release commit, publishedCounts {tests: 1660, skills: 28, hooks: 21}, three platforms.

Marked a prerelease deliberately: the packed-install lane is new and the MLB receipts are deferred. Shipping it as stable 0.2.0 would repeat the failure this work set out to fix — publishing a claim the evidence does not carry.

Promotion was a fast-forward push of dev to main, the repo's established mechanism (gh pr list --base main --state merged returns []).

Residual, stated plainly: no branch or tag protection exists, so the gate is an early warning, not enforcement — a maintainer can still publish by hand through the UI. A tag ruleset is the natural follow-up.