Launch Readiness
A site that builds and deploys is not a site that is ready to launch. A real user-facing product carries deliverables that no test suite checks: a privacy policy the law requires, a favicon and a social preview so it does not look broken when shared, a custom error page instead of the framework's, a cookie banner where tracking demands one, and a security posture that holds when a stranger, not the developer, is the one making requests.
This skill is the consolidated checklist of those deliverables and, for each, how to verify it is genuinely present rather than assumed. It owns none of the depth: where a skill already covers an item, this gate delegates to it and records the result. What it adds is the web-launch items that fall between the engineering skills, and the discipline that no item is marked done without evidence.
1. What this gate is, and what it is not
Three skills sit near this one and are not it:
release-readiness is this revision safe to ship: tests, security
review, migrations, rollback. Gates the code.
production-verification does the deployed system answer a real request.
Gates the deployment.
launch-readiness is the product complete as a deliverable: legal,
discoverability, content, the visible and invisible
things a real launch needs. Gates the product.
They compose. release-readiness clears the code, production-verification
proves it runs, and this gate confirms the product a user meets is whole.
Running one is not running the others.
2. The eight domains
Every item names the skill that owns its depth. This gate does not restate
that skill; it confirms the item is delivered and verified. The full itemised
grid, with a verification for every line, is in
resources/launch-checklist.md. The concrete mapping for a Supabase and
Next.js project, the stack these items are most often written against, is in
resources/supabase-next-appendix.md; the domains below stay stack agnostic.
Legal and compliance
Privacy policy, terms of use, legal notice where the jurisdiction requires
one, a cookie policy, and a cookie consent banner when the site sets any
non-essential cookie or loads any third party that does. The substance of what
personal data is held and why is owned by data-privacy; the document rigor is
owned by administrative-writing.
This gate never writes the legal text. A privacy policy or a terms document carries legal liability, and inventing its wording is the same defect as inventing a legal reference, which the repository forbids. The deliverable here is the page, its structure, the checklist of clauses the site actually needs given the data it handles, the placeholders, and the verification that the page exists, is reachable and is linked. The wording is the owner's, or a lawyer's.
Discoverability
A meaningful title and meta description per page, a social preview image so a
shared link renders as a card rather than a bare URL, a favicon, a
sitemap.xml, a robots.txt that matches the site's real indexing intent, and
canonical URLs. Depth owned by seo-engineering; the favicon and the social
preview image are web-launch items this gate owns directly.
Content and media
Real copy, not placeholder text left in a shipped page. Real images, not framework stock. Every image compressed and sized for the web, and carrying alt text, which is also an accessibility item.
Performance
Page load within a stated budget, images compressed and lazy loaded below the
fold, and a caching strategy that does not serve one user another user's data.
Depth owned by performance-engineering and caching-strategy.
Accessibility
Colour contrast at the required ratio, keyboard operability, and layout that
holds from a phone to a wide screen. Depth owned by accessibility-testing.
Verified by operation, not by a scanner's score alone.
UX integrity
A custom 404 page styled to the project rather than the framework default, no
broken internal or outbound links, forms that validate and report errors
clearly, anti-spam on every public form, one clear primary call to action per
page rather than several competing ones, and no dead button, stub handler or
fake success anywhere a user can reach. The last is owned by
implementation-integrity; the rest this gate owns directly.
Analytics and measurement
An analytics tool actually installed and recording, gated behind consent where
the law requires it, measuring events defined from the questions they answer
rather than everything. Depth owned by analytics-instrumentation, with
data-privacy for the consent boundary.
Security posture
The largest domain, and entirely delegated: this gate runs the security skills
and records their results, it does not re-audit. Secrets absent from the
repository and from the client bundle, no privileged key shipped to the
browser, authorization decided on the server and never trusted from the
client, row level access enforced at the database, sensitive data encrypted,
sessions and passwords handled correctly, login and public endpoints rate
limited and bot resistant, every input validated and every output escaped,
uploads restricted, responses returning only the fields they should, security
headers set, CORS scoped, detailed errors and debug logging kept out of
production, webhooks signature verified, dependencies scanned and current, and
HTTPS enforced. Owned by security-audit, input-validation,
authentication-security, authorization-design, session-security,
security-headers, dependency-security, secrets-management and
file-handling. The itemised grid maps each line to its owner.
Infrastructure and data
HTTPS forced with a valid certificate, email confirmation on signup where
accounts exist, and an automated database backup whose restore has actually
been rehearsed. Owned by deployment-engineering and backup-recovery. A
backup nobody has restored is a hypothesis, not a backup.
3. Verification, per item
Every item resolves to exactly one status, and a status other than
not applicable carries its evidence.
PASS delivered and verified, with the evidence quoted
FAIL absent, or present but broken, with what was observed
NOT APPLICABLE the item does not apply, with the reason
NOT VERIFIED could not be checked in this pass, with what is missing to
check it
PASS is never written from reasoning. A favicon is PASS when the request
for it returned one, not when the tag is in the markup. A backup is PASS
when a restore produced the data, not when the schedule exists. Every PASS
that cannot show its evidence is a NOT VERIFIED instead.
4. Proportionality
Not every site needs every item. A static brochure with no accounts, no forms
and no cookies does not need rate limiting, session security or a consent
banner, and those are NOT APPLICABLE with that reason, not FAIL. The gate
scales to what the product actually is:
has accounts authentication, session, password, login rate limit,
email confirmation all apply
has any form validation and anti-spam apply
sets a cookie consent banner and cookie policy apply, unless every
cookie is strictly essential
holds personal data privacy policy and data-privacy apply
has a database row level access, backup and restore apply
Marking an item NOT APPLICABLE requires the condition above it to be false,
stated. An item is never dropped silently to make the gate pass.
5. Protocol
- Establish what the product is: static or dynamic, accounts or none, forms, cookies, personal data, database. This decides which items apply, per section 4.
- Walk
resources/launch-checklist.mddomain by domain. For each applicable item, delegate to the owning skill where one exists, and check the web-launch items this gate owns directly. - Assign each item a status from section 3, with evidence for anything not
not applicable. - For a Supabase or Next.js project, cross-check against
resources/supabase-next-appendix.md, which names the concrete failure each abstract item maps to on that stack. - Collect every
FAILas a launch blocker and everyNOT VERIFIEDas a gap that must be closed or accepted in writing before launch. - Produce the result: the status per item, the blockers, the gaps, and a
single verdict,
ready,blockedwith the named blockers, ornot verifiedwhen too much could not be checked to judge. - Never issue
readywhile a security-posture item isFAILor a legally required page is absent. Those are not tradeable against a launch date.
6. What this gate refuses
- Writing the legal text of a privacy policy, terms document or cookie policy. It scaffolds and verifies; it does not author liability.
- Marking any item
PASSfrom reasoning rather than observation. - Issuing
readywith an open security-postureFAIL. - Restating the depth of a delegated skill instead of running it.
- Treating a scanner's green score as accessibility or security verification on its own.
- Silently narrowing the checklist to the items that happen to pass.
7. Auto-critique
Score from 0 to 5: every applicable item was actually checked rather than
assumed, each status carries its evidence or its reason, the applicability
decisions in section 4 are justified rather than convenient, no legal text was
fabricated, no security FAIL was waved through, the verdict matches the item
results rather than the launch pressure.
Threshold: no axis below 3, average at least 4. A verdict of ready with any
security-posture FAIL scores 0 overall regardless of the rest, and is
recomposed as blocked.
8. Interfaces
- Upstream:
engineering-core,implementation-integrity,production-verification,release-readiness. - Owns depth jointly with:
seo-engineering,accessibility-testing,performance-engineering,caching-strategy,security-audit,input-validation,authentication-security,authorization-design,session-security,security-headers,dependency-security,secrets-management,file-handling,data-privacy,analytics-instrumentation,deployment-engineering,backup-recovery,administrative-writing. - Downstream:
client-handoverrecords the result in the delivery package,release-engineeringat the release phase. - Run by: the compliance-verifier agent (an agent, named without backticks
here because it is not a skill), and by
delivery-orchestratorat phase 11.