agentleFS
WeaverseWeaverse/.agents/skills/release-verification/SKILL.md

release-verification

Skill0 starsChanged 6 days agoMITView on GitHub

Verify what Cloudflare Workers, Shopify Oxygen and Fly.io actually serve, including cron, queue and release-tag state. Use when shipping to those platforms or checking whether a feature is live; not for preparing version bumps.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the candidate workflow's trigger and conditions for the actual event/ref.

    line 15
  2. Distinctive:

    Check environment configuration by presence and approved digests, never by dumping secret values.

    line 17
  3. Very specific:

    Keep version and changelog in the implementation change before final review/CI.

    line 19
  4. Distinctive:

    Read environment values only through an approved, non-secret projection.

    line 30
  5. Distinctive:

    Keep framework build/codegen, Vite plugins, route preset, worker entry and the supported deploy path intact during toolchain changes.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Release verification

Before cutover

  • A pipeline without a schema step can deploy code against missing columns while CI stays green. Verify the target catalog, not a developer database or generated ORM client. database-migration-review owns schema proof.
  • Compare the candidate with each target's serving source, not just its branch. Manual promotions can contain fixes absent from the branch; a seemingly additive release can roll them back.
  • Read the candidate workflow's trigger and conditions for the actual event/ref. A merge is a separate trigger; changing path filters can itself deploy. Do not assume the previous workflow or a feature-branch exclusion wins.
  • Check environment configuration by presence and approved digests, never by dumping secret values. A green health route may not exercise a newly configured provider.
  • Keep version and changelog in the implementation change before final review/CI. ship prepares a draft; neither a draft nor a development-branch merge is production evidence.

Serving identity

  • Bind producer evidence to consumer readback: intended SHA/version, running platform version or image digest, and a response from the affected route. Match the runtime's version field, not any old changelog text in HTML.
  • Workers can serve multiple versions. Order deployment entries by parsed created_on, inspect the active traffic allocation and all referenced versions, and compare their provenance with the reviewed candidate. An upload ID, message containing a SHA, or row zero from a list is not alone proof of serving identity.
  • For Fly.io, inspect every active serving Machine, its image digest and SHA. A stopped Machine is not serving. Read environment values only through an approved, non-secret projection. Machine config updates may merge env rather than remove absent keys; verify removal explicitly without exposing values.
  • A container COPY . . after generation can overwrite the generated client with local output. Compare .dockerignore and generation order, then inspect the running artifact. A host build or --no-cache is not proof.
  • Peel a remote release tag to its commit and compare with the reviewed/deployed SHA. target_commitish may be a moving branch; an annotated tag-object ID is not a commit ID. Do not repair a mismatch by silently moving tags.
  • WAF or bot-policy refusals can occur before application execution. Report the blocked probe, not an app outage, and do not spoof identity or bypass the policy to complete verification.

Cloudflare schedules and queues

  • Version upload/deploy does not configure cron schedules. Read the live schedules and intended environment separately. A normal deploy can synchronize configuration too; inspect its effects instead of assuming it is code-only.
  • Prove scheduled work in stages: trigger, execution, durable effects, report delivery. A completion marker with zero affected rows is a successful empty run. Empty logs need a known control request and retention window.
  • Analytics can lag; distinguish late data from absent execution. Use a bounded read-only follow-up, not another invocation of a side-effecting job. A self-restarting job needs outcome verification from outside its runner.
  • Healthy HTTP plus a green deploy does not prove queue delivery. Read the queue API's delivery_paused and consumer configuration; list-table output and a queue existing are insufficient.
  • Idempotent cutover automation must tolerate successful DELETE responses with empty bodies. After any partial failure, re-list and reconcile only missing operations. A timed-out write can still apply; read before retrying.

Hydrogen on Oxygen

  • Keep framework build/codegen, Vite plugins, route preset, worker entry and the supported deploy path intact during toolchain changes. Use version-matched Hydrogen instructions instead of reconstructing the workflow.
  • Oxygen deployments are immutable: environment changes require redeployment. Keep routability verification on; a branch named dev is not automatically an isolated preview environment.
  • Probe a locale route, product/collection, CMS content, assets, redirects, robots/sitemap and logs alongside the homepage. Cart-to-checkout and account initiation need authorized test scope; never place a purchase as a probe.
  • A resolved submit or HTTP 200 can still carry errors or userErrors. Assert the intended cart mutation. Correct Storefront query context does not ensure a fresh cart's buyerIdentity.countryCode; SDK mutations can create carts implicitly. An unprefixed API request from a locale page can select the default market.
  • Public locale, translation bundle, CMS locale and provider market are distinct. Verify currency provenance; never relabel an external amount with the route currency. Prefer same-response SSR/hydration consistency over literals that a content edit can invalidate.
  • For deploy-memory problems, inspect the pinned CLI's resolved sourcemap options. A Vite setting that the deploy command overrides is a placebo; measure the production command before changing heap limits.

Readback after a failed wrapper

  • A watcher timeout does not fail the remote run. Save its exit before logging and re-read that same run ID in the correct repository. A trailing echo must not convert failure to success.
  • With pipefail, head/SIGPIPE or a no-match grep can fail a wrapper after a successful mutation. Separate mutation and readback, and settle remote state before retrying. Quote workflow input arguments.
  • An always-run artifact upload can turn a successful cutover red when an expected file is missing. Check live state before rollback. Exclude environment snapshots and provider bodies from uploaded evidence.

Report serving SHA/version, tag identity, verified feature probes, schedule/queue state and unproven paths. "No new failures" on an already-red baseline is not "all gates passed".

Download now (ZIP)

Does it work?

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

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.