Repository navigation
STE-23: landing page — one scroll from the problem to check-it-yourself - #61
Merged
Merged
Conversation
…from copy STE-23 requires claims accurate to what is actually live. A number typed into the copy is accurate exactly once: it goes wrong silently the first time the registry grows, and by then it is a false claim on a public page aimed at people who cannot check it. So liveStats derives every figure from GET /skills at render time and the rug pull from two /check reads. When the API does not answer, the page falls back to a snapshot that carries its own date and a live:false flag, so it can label itself stale. A stale number labelled stale is honest; a stale number presented as live is not. The one trap worth naming: chain_total. The API hides test namespaces by default and CLAUDE.md forbids hiding them silently, so if an older API build omits chain_total, falling back to `total` would quietly assert that nothing is filtered -- the single claim this page must never make by accident. It falls back to the verified snapshot figure instead. SkillList gains chain_total and hidden_test_entries as optional fields. The API has always returned them; the type simply did not say so. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
9 cases. Two failures matter more than the rest, and both are covered explicitly: claiming the chain holds only what we chose to show, and printing a stale number without labelling it. The third case is the one that would slip through review -- an API build that omits chain_total must not cause the page to imply nothing is filtered. The rug-pull tests pin the other direction: if the chain ever stops saying SAFE for v1 and DANGEROUS for v2, the page must fall back rather than narrate a story the chain contradicts. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
One scroll: hero, the problem with its two CVEs, the five-step loop, why Stellar in one sentence, live evidence, and a close. Colours, type and spacing come only from the STE-7 tokens. The hero is not a headline over a gradient. The most characteristic object in this product's world is a skill's identity, and the thesis is that identity is the bytes rather than the name -- so the hero is the card an agent looks at before installing, and the reader flips it between two versions of one real skill. The name does not move a pixel. The hash and the verdict do. That is CVE-2025-54136 as an object instead of a paragraph: Cursor remembered approval under an MCP entry's NAME, so swapping what it ran needed no new prompt. The hash is set as display type because it is the thing being argued about. Numbering the five steps is justified here: it is a real sequence and the order carries information. Verdict is never colour alone -- the card pairs colour with an icon and the verdict word, as brand.md requires. Streaming, not awaiting. GET /skills costs about 2.8s against the live API, so awaiting both reads up front gave a 3.5s blank page, measured in a production build. This is the page we send to people who have never heard of us. Behind Suspense boundaries the prose paints in 11ms and the figures arrive when the chain answers -- same measurement, same build. Caching was the obvious alternative and was rejected: the shared client sets cache: "no-store" deliberately (api-spec §4, a stale verdict is the worst thing this product can serve), and Next 16's `use cache` needs cacheComponents enabled app-wide, which would weaken that guarantee for the dashboard too. The dashboard CTA is a label, not a link, because the dashboard still redirects to Vercel SSO (STE-26). A call to action that lands on a login wall is worse than an honest "coming soon". Route is /landing, not /. Taking / would move Ancung's registry and rewrite the nav before this page has been reviewed; the ticket asks for routing to be agreed with him first, and the flip is one small commit when it is. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Nothing on that page was written from memory. Both CVEs were checked against their published advisories before going on a public page (CVE-2025-54136, CVSS 8.8, Cursor, approval bound to a name; CVE-2025-6514, CVSS 9.6, mcp-remote, first documented RCE from connecting to an untrusted MCP server). The x402 and USDC SAC wording was verified against Stellar's own docs via MCP Stellar Raven during STE-7. The rug pull was read off the live API: v1 SAFE 100 VERIFIED, v2 DANGEROUS 10 refused, different hashes. Two things are recorded as NOT verified rather than left implied: the 375px layout, because the browser harness would not apply a window resize, so mobile rests on the responsive classes and deserves an eyeball in devtools; and the "Instawards" spelling, which is the open question from STE-7 and appears on this page too. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
app.sterish.xyz went public on 24 September (docs/deployments.md, verified HTTP 200). The hero still said "Dashboard — coming soon", which STE-23 allowed only until STE-26 was live. That condition has expired, so the label was now simply false on a page whose whole argument is that claims should be checkable. The closing CTA moves to the dashboard too. It pointed at the raw /skills JSON, which was the honest answer while nothing else could be opened; for a human reader arriving from a landing page, the dashboard is the better destination. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Landing page at
/landing. Merged as a skeleton at Nabil's request — the structure andthe claims are real, but copy and visuals are expected to keep moving.
cc @axel @fable — review welcome after the fact on this one; it was merged early on purpose
so the rest of the team has something to build against.
What it is
One scroll: hero, the problem with its two CVEs, the five-step loop, why Stellar in one
sentence, live evidence, close.
The hero is not a headline over a gradient. The most characteristic object in this product's
world is a skill's identity, and the thesis is that identity is the bytes rather than the
name — so the hero is the card an agent looks at before installing, and the reader flips it
between two versions of one real skill. The name does not move a pixel. The hash and the
verdict do.
That is CVE-2025-54136 as an object rather than a paragraph: Cursor remembered approval under
an MCP entry's name, so swapping what it ran needed no new prompt.
Every number is read, not typed
liveStats.tsderives all figures fromGET /skillsat render time, and the rug pull fromtwo
/checkreads. A number typed into copy is accurate exactly once, then goes wrongsilently — on a public page aimed at people who cannot check it.
The trap worth naming: if an older API build omits
chain_total, falling back tototalwould quietly assert that nothing is filtered. That is the one claim this page must never
make by accident (CLAUDE.md forbids hiding the test namespaces silently), so it falls back to
a dated snapshot instead. Covered by test.
When the API does not answer, the page still renders and labels itself stale.
Streaming, not awaiting
GET /skillscosts ~2.8s against the live API. Awaiting both reads up front gave a 3.5sblank page, measured in a production build. Behind Suspense boundaries the prose paints in
11ms and the figures arrive when the chain answers — same measurement, same build.
Caching was rejected rather than overlooked: the shared client sets
cache: "no-store"onpurpose (api-spec §4 — a stale verdict is the worst thing this product can serve), and Next
16's
use cacheneedscacheComponentsapp-wide, which would weaken that guarantee for thedashboard too.
Claims verified before publishing
Record:
docs/evidence/ste-23-landing-claims-2026-09-20.json.(CVSS 9.6, mcp-remote, first documented RCE from connecting to an untrusted MCP server) —
both checked against their published advisories.
beside x402 in those docs; the page names x402 only.
hashes.
its stellar.expert links.
Updated for today's URLs
app.sterish.xyzwent public on 24 September, so the hero CTA now links to it instead ofreading "Dashboard — coming soon" — a label STE-23 permitted only until STE-26 was live.
Tests
155 passing, 10 files. Typecheck, lint and production build clean. 9 new tests cover the
figures the page states as fact, including the fallback paths and the
chain_totaltrap.One run failed while sharing the machine with
pnpm build(18s vs the usual 2.1s); threeconsecutive clean runs after that.
Known, not hidden
Mobile rests on responsive classes (grids collapse to one column; the 64-char hash and the
skill id carry
break-all). Worth an eyeball in devtools./landing, not/. Taking/moves Ancung's registry and rewrites the nav;sterish.xyzis reserved for this page and currently 404s. The flip is one small commitonce routing is agreed.
Linear
STE-23 — https://linear.app/sterish/issue/STE-23
Unblocks STE-24 (polish pass) and feeds STE-29.
🤖 Generated with Claude Code