prove-its-live
Skill0 starsChanged 22 days agoMITView on GitHub
Before claiming anything is "deployed", "live", "shipped", "in prod" or "fixed in production", prove it - read which commit production actually serves (Vercel API through the CLI's own login, a /version or build-id endpoint, Netlify, Render, Fly, Cloudflare or GitHub Pages) and compare it to what you pushed; optionally move a `production` git tag. Use after every push or deploy, whenever someone asks "is it live?", "what's in production?", "did the deploy go out?", "is my fix up?", and before writing a status update or closing a ticket about production.
- Commits and pushes
In this repository4 other files it gives its agents.
- multi-session-gitSkill
- permission-unblockSkill
- post-deploy-smokeSkill
- secret-guardSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Every sentence you write about production must rest on evidence read in this session: the SHA the platform serves, compared with the SHA …
- Distinctive: line 57
Verify field names against your provider's current API docs.
- Distinctive: line 69
Never write "deployed", "live" or "fixed in prod" from a push message, a CI badge, a dashboard you did not read in this session, or memory.
≈ “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.
Files
Prove it's live
Pushed is not deployed. Deployed is not live. git push printing success says nothing about which
commit your users get. Real failures: the push succeeded but landed on another session's branch, so nothing
deployed; the build failed silently and the old version kept serving; a rollback was promoted and nobody knew.
Every sentence you write about production must rest on evidence read in this session: the SHA the platform serves, compared with the SHA you expected. Otherwise write "not verified".
The check
node ${CLAUDE_SKILL_DIR}/served-sha.mjs --repo /full/path/to/repo${CLAUDE_SKILL_DIR} is this skill's folder. Output, for example:
served : 6c7f852 Product page https://site-abc.vercel.app since 2026-10-25T16:03:13Z
expected : 8f95ca4 Launch posts (origin/main)
not live : ERROR 8f95ca4 https://site-def.vercel.app at 2026-10-25T16:10:02Z
verdict : MISMATCH (behind)Exit 0 = MATCH. Exit 1 = MISMATCH or UNVERIFIABLE. Exit 2 = it could not run (say so; do not guess).
| Flag | Meaning |
--vercel | default when the repo has .vercel/project.json. Reads the deployment the production domains serve (targets.production), not just the newest build, so rollbacks show up. Lists newer production deploys that failed or are still building |
--project <id|name> --team <teamId|slug> | when the repo is not linked |
--url https://site/version.json | read the SHA your build exposes (JSON field sha, commit, gitSha, buildId... or any 40-hex string) |
--sha <sha> | a served SHA you read from another platform (below) |
--ref <ref> | what you expected to be live (default origin/<default branch>, after git fetch) |
--tag / --push-tag | move the production tag to the served commit locally / and push it, so git log --decorate shows what is up |
Vercel auth comes from VERCEL_TOKEN or the Vercel CLI's own login file (vercel login); it is read
in-process and never printed. Never ask the user to paste a token into the chat.
Verdicts and what to do
| Verdict | Means | Do |
| MATCH | production serves exactly the expected commit | now run post-deploy-smoke on the live URLs; only then say "live" |
| MISMATCH (behind) | your commit is not live | check the build log / the not live line; wait or fix; re-check |
| MISMATCH (ahead) | something newer than your ref is live | someone else deployed; fetch and read before saying anything |
| MISMATCH (diverged / unknown) | another branch, or a commit not in your clone | find out who deployed what |
| UNVERIFIABLE | the deployment has no commit SHA (CLI upload without git metadata) | deploy from git, or add a /version endpoint, then check with --url |
Other platforms: get the served SHA, pass it with --sha
Verify field names against your provider's current API docs; these are the usual places.
- Any platform: expose the build commit at a URL (e.g. write
VERCEL_GIT_COMMIT_SHA,COMMIT_REF,RENDER_GIT_COMMIT,GITHUB_SHAinto/version.jsonat build time) and use--url. Most reliable, because it is what the live server actually returns. - Netlify:
netlify api listSiteDeploys --data '{"site_id":"<id>"}'; the published deploy'scommit_ref. - Render:
GET https://api.render.com/v1/services/<id>/deploys?limit=1;commit.idof the live deploy. - Fly.io:
fly releases --json/fly image showshows the image; tag images with the git SHA at build so it is readable. - Cloudflare Pages:
wrangler pages deployment list --project-name <name>; the production deployment's source commit. - GitHub Pages:
gh api repos/<owner>/<repo>/pages/builds/latest --jq .commit. - Your remote, not a platform:
git ls-remote origin refs/heads/<branch>proves what was pushed, not what is deployed.
Rules
- Never write "deployed", "live" or "fixed in prod" from a push message, a CI badge, a dashboard you did not read in this session, or memory.
- Compare against the remote (
origin/<branch>after fetch), not your local HEAD; the script warns when there is no remote. - A MATCH proves the code; it does not prove the page works. Follow with
post-deploy-smoke. --tag/--push-tagonly on repos you own and only when asked; the tag must point at what is served, never at what you hope is served.- Report in one line: served SHA, expected SHA, verdict, and the live URL.
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
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.

