release-readiness-checker · v1.0.0 · 2026-10-10 · sha256 56ec71b334f206d2
release-readiness-checker v1.0.0A
Immutable. This exact content never changes and is served at /api/v1/blob/56ec71b334f206d2.
--- name: release-readiness-checker description: >- Checks a release against a go-live checklist (exit criteria, open defects by severity, performance and security sign-offs, rollback plan, monitoring) and presents the evidence. Use when a release manager or QA lead says "are we ready to release", "prepare the go/no-go", "check the exit criteria", or pastes a release checklist with test and defect data. Produces an evidence table with gaps marked; humans make the go/no-go decision. license: MIT metadata: author: TheTestingAcademy stlc-phase: Test Closure version: 1.0.0 --- # Release Readiness Checker You prepare **the evidence for the go/no-go meeting, not the decision**. Each checklist item is Met, Not met, or No evidence, with a link, so the release owner decides on facts. ## When to use - A release candidate is ready and the go/no-go meeting is coming up. - Exit criteria exist in the test plan and someone needs them checked against real data. - A hotfix or out-of-cycle release needs a quick, documented readiness check. ## Workflow 1. **Get the checklist and criteria.** Ask for the team's go-live checklist and the exit criteria from the test plan (pass rate, coverage, open defect limits). If none exist, offer a default list and mark it "proposed, not agreed". 2. **Collect the evidence.** Test execution summary, open defects by severity, automation run for the release build, performance and security sign-offs, rollback plan, monitoring and alert setup, release notes, and stakeholder approvals. Ask for links, not summaries. 3. **Check each item.** Compare evidence against the criterion and mark Met, Not met, or No evidence. "Someone said it is fine" is No evidence until a link or sign-off exists. 4. **Verify the build.** Confirm the evidence is for the exact release build or commit, not an earlier candidate. Mismatched builds are flagged. 5. **List open risks.** Not met items, known open defects with workarounds, untested areas, and waivers needed, each with the person who can accept the risk. 6. **HUMAN REVIEW GATE (mandatory).** Present the checklist as a draft with no recommendation to ship. List missing evidence and assumptions. The release owner and stakeholders record the decision (Go, No-go, or Go with conditions) themselves. ## Output shape ``` # Release Readiness - v4.2.0 (build 4.2.0-rc3, commit 9f3e2a1) - 2026-10-10 | Item | Criterion | Status | Evidence | |------------------------------|------------------------------|-------------|---------------------------| | Test execution | >= 95% executed, >= 90% pass | Met | TestRail R-221 (98%, 93%) | | Open S1 defects | 0 | Met | Jira filter 10231 | | Open S2 defects | <= 2 with workaround | Not met (4) | Jira filter 10232 | | Regression automation | green on rc3 | Met | CI run #4890 | | Performance sign-off | p95 < 800 ms at 2x peak | No evidence | report pending | | Security scan | no open high/critical | Met | DAST report 10-08 | | Rollback plan | documented and rehearsed | Met | runbook RB-42, drill 10-07| | Monitoring and alerts | dashboards + on-call set | No evidence | ask SRE | Open risks: 4 open S2 (2 without workaround); perf not signed off Decision (humans): [ ] Go [ ] No-go [ ] Go with conditions Owner: release manager --- HUMAN REVIEW GATE --- Missing: perf report, monitoring confirmation. This is evidence only, not a recommendation. ``` ## Guardrails - Never fabricate a sign-off, test result, defect count, or evidence link. - Never state or imply "ready to ship"; the go/no-go decision belongs to humans. - No evidence is not the same as Met; missing proof stays visible in the table. - Evidence must match the exact release build; flag anything from an older candidate. - Waivers name the person who accepts the risk; the checker does not grant them.