Root causes found:
1. syncedUsers blocked re-sync permanently even on failure — first attempt
fires before WebSocket connections are ready on mobile, fails silently,
gets marked done, profile stays anonymous forever that session
2. useAuthor had 5min staleTime — empty {} result from pre-login query
was cached and not refetched after login
3. NostrSync did not force-remove the stale useAuthor cache entry
Fixes:
- Rewrite NostrSync with proper retry logic (up to 5 attempts, backoff)
- Tracks success/failure per pubkey, only stops on confirmed success
- Re-triggers on user.pubkey change (catches Amber polling completing)
- Forces queryClient.removeQueries on author cache before each attempt
- Tries Citrine → bootstrap relays → main pool in order
- useAuthor: reduce staleTime 5min→1min, expand fallback relay list
- Both files now use same BOOTSTRAP_RELAYS set for consistency
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Bug 1 - AppProvider was wiping NIP-65 relay list:
- Relay migration check was comparing stored relays vs defaults and
resetting them if they differed — overwriting the user's own NIP-65 list
- Fix: only reset relays to defaults if updatedAt === 0 (never synced)
and relays array is empty
Bug 2 - NostrSync not finding profile on mobile:
- Add Citrine local relay detection (ws://localhost:4869)
- If Citrine is running, query it first — instant local data
- Add BOOTSTRAP_RELAYS (purplepag.es, relay.nostr.band, nos.lol)
- Always try these for profile+relays before main pool
- syncRelays() now accepts optional relay group param
- Increase initial delay to 1500ms in PWA mode (800ms in browser)
to give WebSocket connections time to establish after Amber redirect
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
PWA standalone mode is single-tab. Opening nostrsigner: with a callbackUrl
sends the callback to the browser (a different app), not back to the PWA.
The PWA tab is suspended and never sees the result.
Fix:
- Detect PWA standalone mode via matchMedia('(display-mode: standalone)')
- In PWA mode: open Amber WITHOUT a callbackUrl — Amber copies pubkey to
clipboard automatically when no callback is given
- Use visibilitychange to detect when user returns to PWA after approving
- Auto-read clipboard on return and complete login silently
- Manual "I approved" button now tries localStorage first (browser), then
falls back to clipboard read (PWA) — covers both contexts with one button
- Browser mode unchanged: localStorage cross-tab polling still works
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Root cause: on GrapheneOS, Amber opens /amber-callback in a BRAND NEW TAB.
login.amber() was being called in that new tab's React instance, which has
no effect on the original tab where the user is waiting.
Fix:
- AmberCallbackPage now writes pubkey to localStorage('amber_pending_pubkey')
instead of calling login.amber() — a cross-tab signal
- LoginDialog polls for 'amber_pending_pubkey' every 400ms and calls
login.amber() itself in the correct React context
- Manual "I approved" button also reads the same key
- Removed reliance on getAmberLogin() which checked the wrong localStorage key
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Both were added in the last mobile optimization and caused touch scroll
to stop working on Android/GrapheneOS browsers:
- overscroll-behavior: none on html/body blocks native touch scroll
- * { -webkit-overflow-scrolling: touch } on every element interferes
with the browser's native scroll handling
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Root cause: window.location.href navigates the tab away entirely on Android,
destroying sessionStorage. When Amber calls back to /amber-callback the login
completes there, but returning to the dialog the state is lost.
Fixes:
- Switch from sessionStorage to localStorage for amber_return_path and
amber_login_attempt_ts — survives tab navigation on Android
- Add 500ms polling loop in handleAmberLogin that checks getAmberLogin()
every 500ms — detects when AmberCallbackPage stored the pubkey and
auto-completes login without needing visibilitychange to fire
- Add "I approved — complete login" manual button in waiting state as
explicit fallback for stuck users (reads localStorage right away)
- AmberCallbackPage now uses localStorage for all persistence
- Clean up poll interval on dialog close and unmount
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- Add dedicated /amber-callback route so Amber always has a stable return URL
- Handles get_public_key (login), sign_event, nip44/nip04 encrypt/decrypt
- Shows animated processing screen while completing, then redirects back
- Add subscribeToAmberReturn() helper in amber.ts using visibilitychange
- Auto-detects when user returns to the tab after cancelling in Amber
- Resets login UI to initial state gracefully
- Update buildCallbackUrl() to use /amber-callback instead of current page URL
- Rewrite Amber tab in LoginDialog:
- Remove 2-step clipboard paste flow entirely
- Single tap opens Amber, login completes automatically on return
- Waiting state shows animated dots + cancel button
- No manual pubkey entry needed
- Save return path in sessionStorage so AmberCallbackPage redirects user back
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- ChatText component renders markdown [label](url) and bare URLs as neon clickable links
- 👉 URL patterns also rendered as clickable links
- get_weather tool: real-time weather + skateability score via Open-Meteo
Ask "weather in Bangkok?" or "is it good to skate in Barcelona today?"
- Footer updated: "Powered by ContextVM & Nostr Relays Data" with ContextVM
linking to https://contextvm.org/s/e3564bb877...
- Weather geocodes city name via Nominatim, returns temp/conditions/wind/skateability
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- Dark forest green + neon lime (#ccff00) theme matching app's THUG2 aesthetic
- Mobile: compact strip above bottom nav, no more full-width overlay
- Minimized: small round button with neon pulse, lives at bottom-left
- Jungle texture strips top/bottom, dark gradient background
- Chat bubbles: neon red for user, neon lime for Sasquatch
- Input styled to match dark app theme
- Neon glow effects, live dot indicator
- Works cleanly on all screen sizes
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The encrypted CVM request/response cycle (kind 21000/21001) was too unreliable
in the browser — timeouts on every call. Replace with direct Nostr relay queries
using nostr.group() which is already proven to work in the app. Same filtering
logic as the server. No dependency on CVM server being up for data queries.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Previous models (llama-3.1-8b, gemma-3-4b, mistral-7b) returned 404 unavailable.
Tested and confirmed working: openai/gpt-oss-20b:free and google/gemma-4-31b-it:free.
Keep llama-3.3-70b as quality option when not rate-limited.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- Try 4 free OpenRouter models in order (rate limit resilience)
- Add greeting responses to static FAQ so hi/hello/hey never need AI
- Fix static FAQ to exact-match short words like "hi" before partial match
- Better error message explaining free tier rate limits
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
nostr-tools v2 SimplePool.subscribeMany() callbacks (onevent/oneose) no
longer fire reliably in browser context, causing the CVM client to always
time out. Switch to raw Relay.connect() which uses the v2 API correctly.
Race across all CVM relays and resolve on first valid response.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The server was using the nostr-tools v1 API (subscribeMany with onevent/oneose
callbacks) but the installed version is v2.23.5 which changed the internal
subscription handling. querySync() is the correct v2 API for one-shot queries
and handles EOSE cleanly without needing manual timeout management.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
CVM server fixes (cvm-server/server.ts):
- Use 'title' tag instead of 'name' to read spot names (app publishes 'title')
- Search now checks 'title' tag (not just 'name') for keyword matches
- City filter now also checks 'title' tag and event content
- Country filter also checks event content
- Features filter now checks 't' tags (app uses 't' not 'feature')
- Validate spots with 't: skatespot' tag (same rule as the app)
- Return 'location' field and proper features list in results
- Image parsing now supports both 'image' (legacy) and 'imeta' (NIP-92)
CreateSpotPage fix (src/pages/CreateSpotPage.tsx):
- Add reverse geocoding via Nominatim on spot submit
- Emit structured 'city' and 'country' tags so CVM can filter by them
- Auto-fill location field from geocoding if user left it blank
SasquatchGuide improvement (src/components/SasquatchGuide.tsx):
- Add data model notes to system prompt explaining tag structure
- Instruct AI to handle 0-result responses gracefully (encourage adding spots)
- Use limit:20+ for city/country searches
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- Rich system prompt that explains the CVM architecture: it aggregates
across multiple relays, deduplicates, and filters by features inside
events (complex queries that plain Nostr filters can't do)
- Switched tool call format from fragile regex (TOOL:name {...}) to
clean JSON {"tool":"name","args":{...}} that's easier for the LLM
- More robust JSON parsing — scans the full AI response for the tool call
- Better second-pass prompt: AI formats results with bullet points + links
- Updated suggested questions to showcase real CVM queries
- CVM failure gives a clear message explaining what the server does
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Added auth.nostr1.com and wotr.relatr.xyz back as default relays.
All 5 relays are now the default for all new users:
- wss://auth.nostr1.com (auth)
- wss://wotr.relatr.xyz (web of trust)
- wss://relay.ditto.pub (general)
- wss://goskatespots.nostr1.com (community - toggleable)
- wss://purplepag.es (profiles - read only)
Updated across: App.tsx, NostrProvider.tsx, NostrSync.tsx, TestApp.tsx
Updated Settings page Default Relays info cards to show all 5 with descriptions.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The 3.1-8b:free model is no longer free on OpenRouter.
Switching to llama-3.3-70b:free which is available and actually
a much better model (70B vs 8B).
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Removed:
- [NostrProvider] Routing query to read relays (fired on EVERY query)
- [NostrProvider] Publishing kind X to relays
- [NostrSync] Starting sync / querying profile / profile query returned / sync completed
- [NostrSync] Profile not found on main relays
- [NostrSync] Profile synced and cached
- [Shakespeare] Fetching / Model / Response status debug logs
Kept all warn/error logs for actual failure cases.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Replaced Shakespeare AI (paid, NIP-98) with OpenRouter free model
(meta-llama/llama-3.1-8b-instruct:free). No login required, no credits
needed - anyone can use Sasquatch AI for free.
- Removed useShakespeare hook dependency from SasquatchGuide
- Added direct callAI() function using OpenRouter API
- AI + CVM live data still works the same way
- Removed login nudge, updated placeholder and AI badge
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
proxy.shakespeare.diy was returning NetworkError on all requests.
Direct calls to ai.shakespeare.diy work fine from a real browser.
Kept debug logging to confirm correct URL and pubkey.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Removed from: useCVMClient, NostrSync, useLoggedInAccounts, useAuthor, SettingsPage
Replaced with relay.damus.io where a fallback was still needed.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Use goskatespots.nostr1.com as primary (matches CVM server config),
plus relay.nostr.band and relay.damus.io as fallbacks.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
All fetch calls to ai.shakespeare.diy now go through proxy.shakespeare.diy
so they work from inside the iframe.diy preview sandbox.
The NIP-98 token is still signed against the original URL (not the proxy
URL) so the server-side signature validation passes correctly.
Affects: sendChatMessage, sendStreamingMessage, getAvailableModels
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Previous attempts used 'shakespeare' and 'claude-sonnet-4-5' — both invalid.
The correct format matches the recentlyUsedModels in ai.json config:
provider/model-name with dots not dashes.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The model name 'shakespeare' does not exist on the API — it was returning:
"Model 'shakespeare' is not supported" (400 invalid_request_error)
Updated both sendChatMessage calls in SasquatchGuide and the default
model in useShakespeare hook (sendChatMessage + sendStreamingMessage).
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The wotr.relatr.xyz relay was generating a flood of AbortError warnings
on every query because it was consistently unreachable/timing out.
Removed from:
- src/App.tsx (default config)
- src/components/NostrProvider.tsx (DEFAULT_RELAYS)
- src/components/NostrSync.tsx (NEW_USER_RELAYS for new users)
- src/test/TestApp.tsx (test config)
Replaced the Settings page entry with purplepag.es (profile aggregator)
which is already in the relay list and actually working.
Updated Sasquatch FAQ answer to reflect the correct relay list.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Previously, the catch block around response.json() would swallow the
actual error body and replace it with a generic "Invalid request" message.
Now:
- Read response as text() first (never fails)
- Log the raw body to console as [Shakespeare API 400] raw body: <text>
- Parse JSON from the text string separately
- Fallback error message includes the raw body so it's visible in console
This will make it easy to diagnose exactly what the Shakespeare API
is rejecting (model name, message format, missing field, etc.)
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
If the AI call throws (400, 401, network error, etc.), Sasquatch now:
- Catches the error silently (logs to console as warning)
- Falls back to the static FAQ answer if one matches
- Falls back to a friendly "try again" message if no FAQ match
- Never shows a crash or blank response to the user
The toolMatch reference inside the try block is properly scoped so the
isLive flag works correctly when the full AI + CVM flow succeeds.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- Add useShakespeare.ts hook (Shakespeare AI via NIP-98 auth)
- Add useCVMClient.ts hook — calls SkateSpots CVM server tools over
Nostr directly from the browser (NIP-44 encrypted request/response,
no @contextvm/sdk bundle needed, pure nostr-tools)
- Rewrite SasquatchGuide to use Shakespeare AI + live CVM data:
* Static FAQ answers for protocol/app questions (fast, no API)
* Shakespeare AI for natural language understanding
* AI detects TOOL: calls in its response and executes them live
* CVM server queried for find_spots, find_events, find_coaches,
find_gear, get_community_stats
* Streaming "thinking..." and "Searching Nostr..." indicators
* Live/offline status badge shows CVM server connectivity
* Graceful fallback when CVM server unreachable
* Login nudge for unauthenticated users
CVM server pubkey: e3564bb877adb0c99a7e4973ccb22e516d7d9eca0027d0bd62a04174ed670530
CVM server host: 185.18.221.2 (LNVPS, Dublin)
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Adds cvm-server/ directory with complete ContextVM server implementation:
- server.ts: Full MCP server with 5 tools (find_spots, find_events,
find_coaches, find_gear, get_community_stats) using @contextvm/sdk
- bootstrap.sh: One-command VPS setup script (Bun, PM2, firewall, keygen)
- skatespots-cvm.service: systemd unit for production deployment
- .env.example: Environment variable template
- DEPLOY.md: Step-by-step deploy guide for LNVPS VPS
- package.cvm.json: Dependencies manifest
The server aggregates data from goskatespots.nostr1.com and other relays,
deduplicates addressable events, and exposes structured JSON tools that
AI agents can discover and call via the ContextVM/Nostr protocol (CEP-6).
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The drawCardToCanvas function had an old version of the card rendering logic
that leaked outside the function body (lines 505-731 were orphaned code).
This code referenced local variables (ctx, p, data, qrX, qrSize, PAD, etc.)
that only existed inside the function, causing ReferenceError at module load time
and crashing the entire app with a blank screen.
Also added missing lucide-react imports (Printer, Loader2, Share2, Download,
Check, Copy, Award, Zap, Info, Palette, Shield, ChevronUp, ChevronDown) that
were used in the JSX but not imported, causing the secondary Printer is not
defined ReferenceError.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The qrcode library internally uses var ctx and canvas.getContext('2d') at
initialization time. In the Shakespeare preview iframe environment, canvas
API is restricted causing "ctx is not defined" console errors on every render.
Fix: replace QRCodeCanvas in the DOM screen preview with a static decorative
QR placeholder (5x5 pixel grid). The real, scannable QR code is still
generated correctly in the exported PNG via QRCode.toDataURL() in the
drawCardToCanvas() async function - this is unaffected by the env restriction
since it only runs when the user clicks Print/Save/Share.
Also remove unused QRCodeCanvas import from RiderCardPage.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Canvas renderer overhaul:
- New 3-zone layout: left sidebar (avatar + stats + since) | right zone
(name/identity/shop/sponsors/skills) | top-right QR block
- Left sidebar has a subtle tinted background with accent-colored separator
- Stats shown as balanced column grid under the avatar
- Shop: "RIDES FOR" label + bold italic Georgia serif in theme shopColor (distinct from accent)
- Sponsors: "SPONSORS" label + regular weight in theme sponsorColor (warm gold/orange)
- Each theme now has shopColor + sponsorColor + sidebarBg + chipBorder fields
- Skills row appears only when skills/badges exist, with accent border on chips
- Full npub (44+remainder chars) shown at bottom of right zone in monospace
- Bottom bar: subtle fill + "SKATE-SPOTS.ORG" left, year right
- Park Access card: same sidebar layout, access level chip in solid accent fill
Share button:
- Uses Web Share API (navigator.share) if device supports it — opens native
share sheet (WhatsApp, iMessage, Telegram, AirDrop, etc.) with PNG file
- Falls back to copying profile URL to clipboard
- Lime green color to visually stand out from Print button
UI cleanup:
- Remove Badges tab entirely — was cluttering the nav
- Award Badge moved inline to secondary button row (yellow, subtle)
- Primary actions: Print | Share | Save PNG (all same size, flex-wrap)
- Secondary actions: Copy npub | Award Badge | Publish to Nostr
- Fixed stray closing divs from tab removal
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
QR fix (critical):
- loadImage() was routing ALL urls through proxy.shakespeare.diy including
data: URIs from QRCode.toDataURL() — proxy rejected them, QR never loaded
- Fix: only proxy http/https URLs; data: and blob: URIs load directly
- QR now renders reliably; scanning opens skate-spots.org/<npub> profile
npub display:
- Split full npub across two lines (32 chars each) instead of truncating
- Both lines shown in monospace below the brand header
- npub is now fully readable on the card
Explicit labeled sections:
- "SKATE SHOP" label in muted above shop name in accent color
- "SPONSORS" label in muted above sponsors text
- Sections only appear if data is entered — no empty space when blank
- Stats (Spots/Zaps/Events) shown as 3-column grid with label+value
Layout improvements:
- QR moved to top-right, rendered before other content (ensures it draws)
- Avatar sized 180px with "SINCE YYYY" underneath
- Identity column flows: name → nip05 → style chip → shop → sponsors → stats → skills
- Bottom divider line + year issued in footer
- fillTextFit() helper for clean text truncation with ellipsis
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
html2canvas was capturing the full browser viewport instead of just the
340×214px card element — causing the giant oversized/wrong layout in the PNG.
New approach: drawCardToCanvas() draws the card entirely via Canvas 2D API:
- Creates an 856×540px canvas (10× credit-card mm dimensions, print-ready)
- Draws gradient background, top stripe, avatar (with CORS proxy), text,
skill chips, stats, and QR code using direct canvas drawing commands
- QR code generated via qrcode lib toDataURL(), then drawn as image on canvas
- Avatar loaded via proxy.shakespeare.diy for CORS, with 4s timeout fallback
- Theme colors from THEME_PALETTE map — same 6 themes as the screen card
- No DOM capture, no scaling issues, no CSS stripping — output is pixel-perfect
QR code explanation:
- Encodes https://skate-spots.org/<npub>
- Scanning opens the rider's SkateSpots profile (NIP19Page handles /npub1...)
- White background patch behind QR for reliable scanning on dark card themes
Other changes:
- "Rider License" → "Skater License" everywhere
- Remove html2canvas dependency usage (kept package but unused for now)
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Problems fixed:
- Card was rendering too tall: replaced flex-1/justify-between layout with
strict pixel-height rows (row1: 26px header, row2: 118px body, fixed bottom bar)
— content no longer floats or overflows
- No QR code: added QRCodeCanvas (existing component) in bottom-right of card
body, encoding https://skate-spots.org/<npub> — scanning takes you to the
rider's profile for verification
- License ID removed: replaced SS-XXXXXXX with shortNpub (npub1xxx...yyyy)
in top-right. The npub IS the identity — no made-up number needed
- Added crossOrigin="anonymous" on avatar img for html2canvas CORS handling
- Updated html2canvas onclone to lock card to 340×214px during capture
- Download filename uses npub prefix instead of licenseId
- verifyUrl() helper generates skate-spots.org/<npub> for QR content
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
