agentleFS
openclawopenclaw/openclaw/.agents/skills/verify-release/SKILL.md

verify-release

Skill394k starsChanged todayMITView on GitHub

Verify regular or extended-stable OpenClaw releases against the exact publication surfaces, workflow identities, package provenance, and published CLI smoke tests expected for that release track.

We split this skill into the rules it gives an agent

16 rules. 2 are common, 3 are distinctive, and 11 are very specific to this project.

  • 2

    Common

    Most of the 8 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 11

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this when asked whether an OpenClaw release is fully released, published, promoted, smoke-tested, or live-verified.

    line 8
  2. Common:

    Verify live state. Do not trust local checkout state, release notes, or old memory as current truth.

    line 20
  3. Very specific:

    If the checkout is dirty or divergent, use it only for scripts/reference.

    line 22
  4. Very specific:

    Keep the final terse: yes/no, evidence bullets, caveats, cleanup.

    line 26
  5. Distinctive:

    Use these checks only for the regular orchestrated release track.

    line 30
  6. Very specific:

    Do not count raw publish flags: openclaw.build.bundledDist === true explicitly defers external publication even when publish flags are set.

    line 56
  7. Common:

    Check the Plugin ClawHub Release workflow conclusion and publish job count.

    line 67
  8. Very specific:

    Use OpenClaw itself for live registry proof: openclaw plugins search <known-plugin> --json.

    line 68
  9. Very specific:

    Prefer matrix unless that plugin is not in the expected set.

    line 73
  10. Very specific:

    Verify conclusions for release notes evidence links: Full Release Validation, OpenClaw Release Checks, OpenClaw NPM Release, Plugin NPM …

    line 76
  11. Very specific:

    Docker: verify exact default, slim, browser, and architecture images and attestations in both registries.

    line 114
  12. Very specific:

    Recovery: never republish. Use promote_extended_stable in the openclaw/releases dist-tag workflow for the root selector (an unsuffixed …

    line 121
  13. Very specific:

    After the track-specific publication checks pass, use a disposable installation/cache with isolated OpenClaw state:

    line 130
  14. Very specific:

    Run at least one harmless command that touches the published CLI surface, for example plugins --help or gateway --help.

    line 134
  15. Distinctive:

    Do not start an extra local Gateway or run an ad hoc live model turn as a preliminary release check.

    line 137
  16. Distinctive:

    Run additional manual live checks only when explicitly requested.

    line 140

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Verify Release

Use this when asked whether an OpenClaw release is fully released, published, promoted, smoke-tested, or live-verified. This is a verification skill, not a publish skill; use $release-openclaw-maintainer before changing release state.

Rules

  • Resolve short suffixes like .27 to the concrete CalVer version from the current date/context, then say the resolved version.
  • Resolve the track first. Both tracks use the shared GitHub Release evidence ledger. Regular beta/stable also uses the platform graph; extended-stable uses its canonical branch, npm selector, and Gateway surfaces. Do not require one track's native or ClawHub artifacts from the other.
  • Verify live state. Do not trust local checkout state, release notes, or old memory as current truth.
  • If the checkout is dirty or divergent, use it only for scripts/reference. For version metadata, fetch from GitHub release/tag or unpack the tag tarball under /tmp.
  • Never print secrets.
  • Keep the final terse: yes/no, evidence bullets, caveats, cleanup.

Regular beta/stable checks

Use these checks only for the regular orchestrated release track.

  1. GitHub release:
    • gh release view v<VERSION> --repo openclaw/openclaw --json tagName,name,publishedAt,isDraft,isPrerelease,targetCommitish,url,body,assets
    • Confirm stable releases are not draft/prerelease.
    • Confirm release body has npm, CI, plugin npm, ClawHub, mac/appcast evidence links when expected.
    • Confirm assets expected for stable mac releases are uploaded: zip, dmg, dSYM, dependency evidence, immutable full-validation manifest, postpublish evidence, and stable-main closeout manifest.
    • Download each immutable evidence asset and its .sha256 companion, then verify the checksum before trusting the release record.
  2. Root npm:
    • npm view openclaw@<VERSION> version dist-tags.latest dist.tarball dist.integrity time.<VERSION> --json
    • latest must equal <VERSION> for stable.
    • Record tarball, integrity, publish time.
    • Confirm the release postpublish evidence records npmRegistrySignaturesVerified: true and npmProvenanceAttestationMatched: true.
  3. Plugin publish set:
    • Get exact tag metadata from GitHub, not the local checkout when dirty: download https://api.github.com/repos/openclaw/openclaw/tarball/v<VERSION> into /tmp/openclaw-v<VERSION>-src.
    • Derive the full expected npm and ClawHub package sets for the release track with the canonical publication planners/collector from the recorded release Tooling SHA, using the exact tag's package metadata. Do not count raw publish flags: openclaw.build.bundledDist === true explicitly defers external publication even when publish flags are set. Record deferred package names and reasons separately.
    • Reconcile expected package identities, versions, and counts across original publication, previously published versions, and selected recovery runs using immutable publication plans, registry readback, and workflow jobs. A selected recovery subset must not narrow the full expected release set: gh api repos/openclaw/openclaw/actions/runs/<RUN>/jobs --paginate.
    • Each expected npm plugin must have version <VERSION> and dist-tags.latest === <VERSION>.
  4. ClawHub:
    • Check the Plugin ClawHub Release workflow conclusion and publish job count.
    • Use OpenClaw itself for live registry proof: openclaw plugins search <known-plugin> --json.
    • Install one official plugin at the exact requested release version from ClawHub in an isolated HOME: openclaw plugins install clawhub:@openclaw/matrix@<VERSION>. Prefer matrix unless that plugin is not in the expected set. ClawHub versions belong in the spec; --pin is only supported for npm installs.
  5. Release workflows:
    • Verify conclusions for release notes evidence links: Full Release Validation, OpenClaw Release Checks, OpenClaw NPM Release, Plugin NPM Release, Plugin ClawHub Release, mac preflight/validation/publish when stable mac assets are expected.
    • For stable, verify OpenClaw Stable Main Closeout succeeded and its manifest records the matching release tag, current rollback drill, stable soak, and blocking performance evidence.
    • Summarize only relevant successful/failed jobs; ignore routine skipped optional lanes unless the release body promised them.

Extended-stable checks

Extended-stable has a GitHub Release with shared release evidence but no native or ClawHub artifacts. Verify it alongside the live tag, workflow, registry, provenance, and image state.

  1. Identity: require final v<VERSION> at patch 33+, with no suffix, contained in extended-stable/YYYY.M.33. Only an active candidate must equal the tip. Root and every publishable official plugin must declare <VERSION>. Require the Git tag and a public, non-prerelease GitHub Release whose title and canonical body match the tag. Require isLatest=false, the dependency evidence, immutable Full Release Validation manifest, postpublish evidence, and their checksums. Require no native or ClawHub assets.
  2. Workflow chain: find the successful parent release run plus its preflight, complete validation, plugin npm, and core publish children. Require a protected release-publish/* parent and canonical release-ci/* validation producer with verified workflow SHA provenance. Validation must use rerun_group=all, release_profile=stable, blocking soak/performance, and the saved attempt. Core publish must reference all three run IDs and bind its manifest, workflow ref, and tarball digest to the release SHA.
  3. Registry: require exact and extended-stable selectors to return <VERSION> for root, every preflight corePackageTarballs entry, and every publishToNpm === true official plugin derived from the tag. Compare the plugin plan, jobs, and complete readback; never infer inventory from diffs.
  4. Provenance: from trusted current tooling, run node --import tsx scripts/openclaw-npm-postpublish-verify.ts <VERSION>. Require signatures, canonical-branch provenance, and publish/preflight digest binding to the release SHA. Preserve output and workflow URLs.
  5. Docker: verify exact default, slim, browser, and architecture images and attestations in both registries. Only the three extended-stable* aliases may resolve to those digests. Require the successful OpenClaw Release Publish parent run and its completed Docker verification. The normal route finalizes afterward; an explicitly requested fast path may activate GitHub first. Repair aliases through current-main Docker Channel Promotion for the exact tag, without rebuilding.
  6. Recovery: never republish. Use promote_extended_stable in the openclaw/releases dist-tag workflow for the root selector (an unsuffixed final patch 33+) and approved credential-isolated tooling for others, then repeat complete readback. Do not require ClawHub, native/mobile apps, website, private dist-tags, or regular latest. Require shared release evidence, but do not require regular native or ClawHub assets.

Published package smoke

After the track-specific publication checks pass, use a disposable installation/cache with isolated OpenClaw state:

  • npm exec --yes --package openclaw@<VERSION> -- openclaw --version.
  • Run at least one harmless command that touches the published CLI surface, for example plugins --help or gateway --help.

Do not start an extra local Gateway or run an ad hoc live model turn as a preliminary release check. Use the canonical release-validation workflows and publication evidence above; this does not remove any of their required lanes. Run additional manual live checks only when explicitly requested.

Caveats To Report

  • Dist-tag caveat: stable latest is release truth; if optional beta mirrors still point at a beta version, report it as a caveat, not a stable-release blocker, unless the user asked to verify beta promotion.
  • Track caveat: name the track and intentionally absent surfaces. Do not call missing regular-release artifacts an extended-stable failure.
  • Divergent checkout caveat: say when local source SHA differs from release tag or origin and which live sources were used instead.

Download now (ZIP)

Trending

Less talk lately.about the whole repository

Dec 2025: 1 mentionJan 2026: 12 mentionsFeb 2026: 27 mentionsMar 2026: 13 mentionsApr 2026: 5 mentionsMay 2026: 8 mentionsJun 2026: 4 mentionsJul 2026: 8 mentionsAug 2026: 3 mentionsSep 2026: 7 mentionsOct 2026: 4 mentionsNov 2026: 0 mentions
Dec 2025Jun 2026Nov 2026
Mentions a month of the openclaw/openclaw repository that our weekly crawl found. There is no adoption data for this file yet.

Who's talking about openclaw/openclaw

The biggest 5 of 92 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 92

Talked about on Bluesky, GitHub, the web and Hacker News (92 times).

On Bluesky · 48

43 more on Bluesky

On GitHub · 17

12 more on GitHub

On the web · 15

10 more on the web

On Hacker News · 12

7 more on Hacker News

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.