Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

chore(prompts): file orders 419-420 for sitemap duplicate entries and a hover-state regression

nirholasae1c2d66e15e
diff --git a/prompts/finish/419-auto-sitemap-duplicate-entries.md b/prompts/finish/419-auto-sitemap-duplicate-entries.md
new file mode 100644
index 000000000..b8ebe5ea1
--- /dev/null
+++ b/prompts/finish/419-auto-sitemap-duplicate-entries.md
@@ -0,0 +1,102 @@
+# 419. Fix duplicate /grok and /docs/creator-payouts sitemap entries, and the noindex page submitted in the sitemap
+
+How to run: paste this file's repo path into a fresh Claude Code chat in this repository and say "run this work order". Runnable now, no gate, no owner action required.
+
+## Operating clause (binding)
+
+- Read CLAUDE.md first. CLAUDE.md overrides everything, including this file.
+- Finish 100% in this session. Never end the turn with a question, an option list, or an unexecuted plan.
+- Bug fixes are test-first: the failure is reproduced below. Confirm it fails for the stated reason, fix it, confirm green, commit test and fix together.
+- Concurrent agents share this worktree: stage explicit paths only (never `git add -A` or `git add .`), re-check `git status` and `git diff --staged` before committing.
+- No em-dash/en-dash anywhere you write.
+
+## Step 0: re-derive the current state
+
+ npm run audit:seo
+
+As measured 2026-10-24, this prints `ERROR (6)` including:
+
+ title-duplicate × 3
+ https://three.ws/grok: 2 pages share "three.ws for Grok Bot: a free 3D studio and a body for Grok · three.ws": https://three.ws/grok, https://three.ws/grok
+ https://three.ws/trenches: 2 pages share "Trenches Pulse · three.ws": https://three.ws/trenches, https://three.ws/docs/trenches
+ https://three.ws/docs/creator-payouts: 2 pages share "Docs · Creator payouts · three.ws": https://three.ws/docs/creator-payouts, https://three.ws/docs/creator-payouts
+ description-duplicate × 2
+ https://three.ws/grok: 2 pages share one description: https://three.ws/grok, https://three.ws/grok
+ https://three.ws/docs/creator-payouts: 2 pages share one description: https://three.ws/docs/creator-payouts, https://three.ws/docs/creator-payouts
+ noindex-in-sitemap × 1
+ https://three.ws/irl/sign: robots: noindex, follow
+
+Confirm the root cause directly in the data file:
+
+ grep -n '"path": "/grok"' data/pages.json
+ grep -n '"path": "/docs/creator-payouts"' data/pages.json
+ grep -n '"path": "/irl/sign"' data/pages.json
+ curl -s https://three.ws/sitemap/core.xml | grep -c '<loc>https://three.ws/grok</loc>'
+ curl -s https://three.ws/sitemap/core.xml | grep -c '<loc>https://three.ws/docs/creator-payouts</loc>'
+
+As measured 2026-10-24: `/grok` appears at `data/pages.json` lines 664 and 6647 (two different entries, `added: "2026-10-08"` vs `added: "2026-10-09"`, the second a richer rewrite that was never deleted from the first add). `/docs/creator-payouts` appears at lines 7286 and 7294, byte-for-byte identical (same title, description, priority, changefreq, `added: "2026-10-23"`), a plain double-add. Both print `2` from the `grep -c` on the live sitemap, confirming the duplicate rows really do produce two `<loc>` entries for the same URL in `sitemap/core.xml`. `pages/irl-sign.html` line 12 carries `<meta name="robots" content="noindex, follow" />` (a deliberate choice: `/irl/sign` is a per-agent printable utility page, not a destination worth ranking), yet its `data/pages.json` entry (line 305) still has it submitted via the sitemap with a real `priority`/`changefreq`, which tells Google to crawl and index a URL the page itself says not to index.
+
+If step 0 now prints 0 errors, run `git log --oneline -5 -- data/pages.json` to find the fix, confirm the three duplicates/noindex-conflict are gone, and retire this order per the "Never blocked" table.
+
+## Root cause
+
+`data/pages.json` is the single source of truth `scripts/build-sitemap.mjs`-equivalent sitemap/pages tooling reads to emit `public/sitemap/core.xml` and feed `audit:pages`/`audit:seo`. Nothing in the add-a-page workflow checks for an existing entry with the same `path` before appending a new one, so:
+1. A page that gets a copy refresh (as `/grok` did between 2026-10-08 and 2026-10-09) can get a second entry instead of an edit to the first, if whoever added the refresh appended rather than searched.
+2. A page added twice by mistake (as `/docs/creator-payouts` was on 2026-10-23) produces an exact duplicate.
+3. Nothing cross-checks a page's own `<meta name="robots">` against its sitemap inclusion, so a page that opts itself out of indexing (`/irl/sign`) still gets submitted.
+
+## Task
+
+1. Open `data/pages.json` and remove the stale `/grok` entry at line 664 (`added: "2026-10-08"`, title `"three.ws for Grok"`), keeping the 2026-10-09 entry (line 6647, the fuller "three.ws for Grok Bot..." copy, `changefreq: "weekly"`).
+2. Remove one of the two identical `/docs/creator-payouts` entries (lines 7285-7300), keeping a single copy.
+3. Remove the `/irl/sign` entry (lines 304-311) from `data/pages.json` entirely: a page whose own HTML opts out of indexing must not be submitted through the sitemap. Do not change `pages/irl-sign.html`'s `noindex` meta tag; that is the correct, deliberate choice for a printable per-agent utility page.
+4. Search `data/pages.json` for any other exact-duplicate `path` values beyond these three, since the audit only reports title/description collisions (which catches duplicates that share copy, but would miss a hypothetical duplicate pair that was independently given two different titles):
+
+ node -e "
+ const pages = require('./data/pages.json');
+ const flat = [];
+ (function walk(node) {
+ if (Array.isArray(node)) { node.forEach(walk); return; }
+ if (node && typeof node === 'object') {
+ if (typeof node.path === 'string') flat.push(node.path);
+ Object.values(node).forEach(walk);
+ }
+ })(pages);
+ const seen = new Map();
+ for (const p of flat) seen.set(p, (seen.get(p) || 0) + 1);
+ for (const [p, n] of seen) if (n > 1) console.log(n, p);
+ "
+
+ Fix any additional duplicate the same way (keep the most recent/complete entry, remove the rest).
+5. Rebuild the sitemap and page-index artifacts that read from `data/pages.json`:
+
+ npm run build:pages
+
+6. Re-run `npm run audit:seo` and confirm `title-duplicate`, `description-duplicate` and `noindex-in-sitemap` no longer appear in the `ERROR` section.
+7. Add a regression guard so a future duplicate `path` in `data/pages.json` fails fast instead of silently shipping to the sitemap: extend `scripts/audit-page-index.mjs` (run via `npm run audit:pages`, already in `npm run gate`) to flag any `path` value that appears more than once across `data/pages.json`, with a clear error naming the duplicate path and both source locations (line numbers if practical, otherwise the duplicate count). Add a corresponding test case or fixture under `tests/` (follow that script's existing test file if one exists, e.g. `tests/audit-page-index.test.js`; create one matching the project's existing audit-script test pattern if it does not) that feeds a pages array with a deliberate duplicate `path` and asserts the audit reports it.
+8. Re-run `npm run audit:pages` and confirm it is still green against the de-duplicated `data/pages.json`, and that the new test from step 7 fails against a duplicate fixture and passes once the duplicate is removed.
+
+## Done when
+
+- `npm run audit:seo` prints 0 errors in the `title-duplicate`, `description-duplicate` and `noindex-in-sitemap` categories.
+- `curl -s https://three.ws/sitemap/core.xml` is not checkable until deploy, but `public/sitemap/core.xml` (regenerated by `npm run build:pages`) contains exactly one `<loc>` entry each for `https://three.ws/grok` and `https://three.ws/docs/creator-payouts`, and zero for `https://three.ws/irl/sign`.
+- `data/pages.json` has no duplicate `path` values anywhere (step 4's one-off script prints nothing).
+- `scripts/audit-page-index.mjs` now rejects a duplicate `path`, proven by a new passing test.
+- `npm run audit:pages`, `npm run audit:docs` and `npm run test:core` show no new failures versus the pre-change baseline.
+- `git diff` touches only: `data/pages.json`, `public/sitemap/**` (and any other `build:pages` output, e.g. `public/sitemap-index.xml`, `data/pages-search.json` if that exists), `scripts/audit-page-index.mjs`, and the new/updated test file.
+- Committed on `evolve` with an explicit-path commit naming the fix.
+- This file is removed, with a dated entry added to `prompts/finish/_context/production-100-PROGRESS.md` citing the commit SHA.
+
+## Never blocked
+
+| Looks like a blocker | It is not | Do this |
+|---|---|---|
+| Unsure which of the two `/grok` entries to keep | The 2026-10-09 one is strictly fuller copy with a tighter `changefreq`; the 2026-10-08 one is the stale first draft | Keep 2026-10-09, remove 2026-10-08 |
+| Tempted to "fix" `/irl/sign` by removing its `noindex` meta tag instead | That tag is deliberate: it is a printable, parameterized per-agent utility page with no standalone SEO value, and indexing it would create near-duplicate-content bloat across every agent's printed sign | Remove the `data/pages.json` entry instead; leave `pages/irl-sign.html` untouched |
+| `npm run build:pages` regenerates more files than expected and the diff looks large | The script is deterministic over the whole `data/pages.json`; a large diff on first run after a multi-entry edit is expected, not a sign something broke | Confirm the diff is only removed/shifted entries (no content corruption), not a reason to abort |
+| Step 0's counts are already 0 when you check | Another agent may have shipped this | Confirm via `git log --oneline -5 -- data/pages.json`, verify the three issues above are gone, retire this order without further changes |
+| `/trenches` vs `/docs/trenches` sharing one title (`title-duplicate` row 2) looks like the same bug | It is a different, lower-priority copy issue: two genuinely distinct pages with coincidentally identical titles, not a duplicate sitemap row | Out of scope for this order; leave it unless you have time left after steps 1-8, in which case retitle `/docs/trenches` to `"Docs · Trenches Pulse"` matching the `/docs/creator-payouts` naming convention and note it in the report |
+
+## Report format
+
+State: the step-0 "before" `audit:seo` error list, the exact lines removed from `data/pages.json` for each of the three issues, the step-4 duplicate-scan output (before and after), the `scripts/audit-page-index.mjs` diff and its new test's passing output, the `npm run build:pages` diff summary, the "after" `audit:seo`/`audit:pages`/`audit:docs`/`test:core` results, and the commit SHA.
diff --git a/prompts/finish/420-auto-hover-state-regression.md b/prompts/finish/420-auto-hover-state-regression.md
new file mode 100644
index 000000000..04daf0042
--- /dev/null
+++ b/prompts/finish/420-auto-hover-state-regression.md
@@ -0,0 +1,81 @@
+# 420. Fix the interactive-state regression: two clickable selectors with no hover rule
+
+How to run: paste this file's repo path into a fresh Claude Code chat in this repository and say "run this work order". Runnable now, no gate, no owner action required.
+
+## Operating clause (binding)
+
+- Read CLAUDE.md first. CLAUDE.md overrides everything, including this file.
+- Finish 100% in this session. Never end the turn with a question, an option list, or an unexecuted plan.
+- Bug fixes are test-first: the failure is reproduced below. Confirm it fails for the stated reason, fix it, confirm green, commit test and fix together.
+- Concurrent agents share this worktree: stage explicit paths only (never `git add -A` or `git add .`), re-check `git status` and `git diff --staged` before committing.
+- No em-dash/en-dash anywhere you write.
+
+## Step 0: re-derive the current state
+
+ npm run audit:states
+
+As measured 2026-10-24, this fails with:
+
+ ✗ interactive-state gaps increased: 2 clickable selector(s) with no hover rule (baseline 0).
+
+ src/earnings.css: 1 (was 0)
+ .er-toggle
+ public/marketplace.css: 1 (was 0)
+ .sp-bundle > summary
+
+`scripts/audit-interactive-states.baseline.json` currently reads `{"total": 0, "perFile": {}}` (confirm with `cat scripts/audit-interactive-states.baseline.json`), so this is a genuine regression against a clean baseline, not a pre-existing known gap. `scripts/audit-interactive-states.mjs` flags any CSS selector that sets `cursor: pointer` (promising the element is clickable) with no sibling `:hover` rule; this is the automated form of the CLAUDE.md definition-of-done requirement "every interactive element has hover, active, and focus states."
+
+Confirm each site directly:
+
+ grep -n '\.er-toggle' src/earnings.css
+ grep -n '\.sp-bundle' public/marketplace.css
+
+`src/earnings.css` line 134: `.er-toggle { display: inline-flex; align-items: center; gap: 10px; cursor: pointer; font-size: var(--text-sm); }` is a label wrapping a checkbox toggle on the earnings page; line 136 already gives its inner `input` a `:focus-visible` ring, but `.er-toggle` itself has no `:hover`.
+
+`public/marketplace.css` line 5506: `.sp-bundle > summary { cursor: pointer; font-size: 13px; margin-bottom: var(--space-sm); }` is a disclosure `<summary>` for the publish-skill supply-chain scan results; line 5572 already gives it a `:focus-visible` ring (`#sp-check:focus-visible, .sp-bundle > summary:focus-visible { outline: 2px solid var(--accent, #00d9ff); outline-offset: 2px; }`), but no `:hover`.
+
+If step 0 now passes (`✓` / gaps at or below baseline), run `git log --oneline -5 -- src/earnings.css public/marketplace.css scripts/audit-interactive-states.baseline.json` to find the fix and retire this order per the "Never blocked" table.
+
+## Root cause
+
+Both selectors were added (or had `cursor: pointer` added) without a matching `:hover` rule, and nothing caught it before merge because `npm run audit:states` is not wired into a pre-commit hook, only into manual/CI-style runs. The baseline mechanism exists precisely to catch this kind of drift, and it did: `total` went from 0 to 2.
+
+## Task
+
+1. In `src/earnings.css`, add a `:hover` rule immediately after the existing `.er-toggle` block (after line 134), following this file's own token conventions used elsewhere (e.g. `var(--ink-dim)`, `var(--stroke-strong)` as seen in `.er-run`/`.er-confirm` nearby). A minimal, consistent choice: lighten the label text or add a subtle background, for example:
+
+ .er-toggle:hover { color: var(--ink); }
+
+ Adjust to whatever reads correctly against `.er-toggle`'s actual rendered content (a checkbox plus label text) when checked visually; the exact tokens matter less than that a visible, non-zero change occurs on hover.
+
+2. In `public/marketplace.css`, add a `:hover` rule immediately after the existing `.sp-bundle > summary` block (after line 509), following the same file's own precedent for similar clickable summary/button elements (e.g. `.market-sort:hover { background: var(--surface-2); border-color: var(--accent); }` at line 156). A consistent choice:
+
+ .sp-bundle > summary:hover { color: var(--accent); }
+
+ or a background treatment if that reads better against the surrounding `.sp-bundle`/`.sp-scan` block styling; verify visually.
+
+3. Re-run `npm run audit:states` and confirm it reports 0 gaps (back to baseline).
+4. Do not run `--update` on the baseline; the fix should restore the existing 0-total baseline, not raise it to accept the regression.
+
+## Done when
+
+- `npm run audit:states` exits with `✓` (gaps at 0, matching `scripts/audit-interactive-states.baseline.json`'s existing `{"total": 0}`).
+- `scripts/audit-interactive-states.baseline.json` is unchanged (still `{"total": 0, "perFile": {}}`); the fix closes the gap in the CSS, not by relaxing the baseline.
+- Visual check: open `/earnings` and the publish-skill flow (wherever `.sp-bundle` renders, search `src/` for the class to find the page) in a browser or describe the rendered change if a browser is unavailable, confirming the new hover states are visible and not jarring.
+- `npm run test:core` shows no new failures versus the pre-change baseline.
+- `git diff` touches only: `src/earnings.css`, `public/marketplace.css`.
+- Committed on `evolve` with an explicit-path commit naming the fix.
+- This file is removed, with a dated entry added to `prompts/finish/_context/production-100-PROGRESS.md` citing the commit SHA.
+
+## Never blocked
+
+| Looks like a blocker | It is not | Do this |
+|---|---|---|
+| Unsure of the exact hover color/background to use | The audit only checks that a `:hover` rule exists for the selector, not its specific visual treatment | Pick a token-based treatment consistent with nearby rules in the same file (cited above), confirm it renders visibly, move on |
+| Tempted to fix by raising the baseline with `--update` | That locks in the regression instead of fixing it; the baseline exists to catch exactly this drift | Add the missing `:hover` rules instead; only ever lower the baseline (via `--update`) when gaps are proven below it, never to accept new ones |
+| Step 0 already passes when you check | Another agent may have shipped this | Confirm via `git log --oneline -5 -- src/earnings.css public/marketplace.css`, verify both selectors now have `:hover` rules, retire this order without further changes |
+| A browser is not available to visually confirm the hover treatment | Visual confirmation is a nice-to-have, not a blocker on an automated-audit-driven CSS fix | State in the report that the visual check could not be performed and why; the mechanical audit passing is the authoritative done-check |
+
+## Report format
+
+State: the step-0 "before" `audit:states` failure output, the exact `:hover` rule added to each file with its line number, the "after" `audit:states` output (must show `✓`), confirmation the baseline file is unchanged, the `npm run test:core` before/after failure counts, and the commit SHA.