Point SKATE_RELAY, the default relay pools, the DM/CVM/zap relay arrays, the cvm-server + push-service defaults, and the archive default at wss://relay.skate-spots.org, keeping the legacy hosted relay as a fallback during the client-update window. Restored 115/115 archived events; the relay runs behind Caddy with TLS. Docs + AGENTS.md updated.
Every provider lived at /coach/<pubkey>/<id> regardless of craft (a barber got
a "coach" URL). New src/lib/labRoutes.ts maps craft kind → segment and builds
the canonical path; the router adds /lab/:craft/:pubkey/:identifier (the craft
segment is cosmetic — the page still resolves by pubkey + identifier) and keeps
/coach/... working for old links. All generated links updated: offer-form
redirects, Lab cards + "Your Lab profiles", cross-links, spot/event/admin links,
the CVM client URL, the share path, and the SPA clean-link shim.
useCVMClient's findCoaches ROLE_KIND, self-tag set and getCommunityStats stopped
at 35012, so Sasquatch could not find the event-organizer (35013) or barber
(35014) profiles we added. Extend LabRole, ROLE_KIND, SELF_TAGS and the stats
queries/counts (lab_total) to cover all Lab kinds 35001-35014.
- sasquatchBrain: abort each provider request after 15s and give up the whole
multi-provider attempt after 35s, so a hung OpenRouter/Pollinations call can no
longer wedge the chat (it now falls through to the static/local fallbacks).
- SasquatchGuide: send the system prompt as a real system message (both the
first pass and the tool-synthesis pass) instead of smuggling it into a user
turn, and mark tool output as untrusted — blunts prompt injection via event
content.
- useCVMClient: signature-verify (verifyEvent) every relay result, matching the
standalone server and the release-notes claim.
- useCVMServerHealth: stale threshold 1h -> 45m to match the server heartbeat.
Two type errors (surfaced by `npm run test` / tsc --noEmit on a full
toolchain; the in-browser bundler does not run tsc, so they were latent):
- useCVMClient.ts: remove a malformed cast
`spots as typeof spots & { tags; created_at }[]` feeding
deduplicateByIdentity(). The intersection dropped pubkey/kind and broke
the reassignment to `spots`. deduplicateAddressable() already returns
the correct shape, so the cast was both wrong and unnecessary.
- queryPersister.ts: import PersistedClient / Persister from
@tanstack/query-persist-client-core (their canonical source) instead of
@tanstack/react-query-persist-client, whose `export *` re-export tsc
won't resolve for named type imports. Added the core package (5.102.8,
matching the pinned wrapper) as an explicit dependency.
Both are pre-existing issues on the branch, unrelated to the audit
remediation commits.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The old identity dedupe keyed on `name|place` where place was a single field
(geohash OR location OR city, first non-empty). Two copies of the same spot —
one carrying a geohash, one carrying only a city — produced different keys and
both survived (the Yellow Park duplicate).
Now:
- Name is reduced to a signature that strips trailing generic words
(skatepark/park/plaza/spot/diy/...) and leading articles, so "Yellow Park",
"Yellow Park Skatepark" and "Yellow Park " collapse to "yellow".
- Places are compared on human-readable tokens (location/city/country) only;
the opaque geohash is excluded. An empty place on either side is treated as a
wildcard, so a geohash-only copy still merges with a "Bangkok" copy.
- Genuinely different spots (different name signatures, or non-empty places with
no shared token) are still kept separate; single-word names are preserved so
generic names don't over-collapse.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
The local fast-path was stricter than the AI it replaced: a typo like "bangok"
won't substring-match "Bangkok", so the fast-path returned a false "No spots
listed there yet" even though Bangkok has spots.
Two-layer fix:
1. Safety net (most important): if a finder fast-path comes back EMPTY, fall
through to the AI path instead of rendering "nothing found". The AI normalizes
spelling and can still find results. Stats/weather always render immediately.
Net effect: fast when the local match works, correct-always via AI when it
doesn't — never a false empty.
2. Typo tolerance: findSpots city matching now uses placeMatches() — substring
OR edit-distance-1 against candidate word tokens (only for queries ≥4 chars,
so short words can't over-match). So "bangok" matches "Bangkok" on the fast
path too. Freeform content stays a strict substring check to avoid false hits.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Two bugs from "spots in bangkok" (said "Found 20" but showed 6, with repeats):
1. Misleading count: formatLocalResults hard-capped the visible list to 6 while
the headline reported the full count. Now shows up to 10 and appends a
truthful "…and N more — see them all on the map" line, so the number and the
list always agree.
2. Duplicate spots: coordinate dedupe (kind:pubkey:d) can't catch the same
real-world spot re-posted under a different identifier or by another account
(e.g. "Yellow Park – Sukhumvit 50" appearing twice). Added
deduplicateByIdentity() — a second-pass dedupe on a normalized
name + geohash/location signature, keeping the freshest event — applied in
findSpots so the count and list are clean.
Both fixes flow through the AI tool-call path and the offline fallback path.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
Three low-risk speedups focused on the slowest path — questions that hit live
relay data:
1. Relay timeout 10s → 4.5s (6s for the fan-out community stats). The useful
relays answer in ~2s; the old 10s just let one slow/dead relay stall the
whole reply.
2. Results-first: for finder tools (spots/events/coaches/gear/crews/shops/
tours/stats/weather) format the CVM result locally and SKIP the second
"prettify" AI call. Removes a full model round-trip (~2–4s) — users want the
list, not extra prose. New toolCallToIntent() maps a tool call to the existing
formatLocalResults() path. Personal/Pro tools (get_my_*) still use AI synthesis.
3. Faster AI fallback: cap OpenRouter attempts to the top 3 free models before
the keyless backup, and shorten 429 backoff (≤800ms vs ≤2500ms). Avoids the
worst-case 15–20s grind through 8 models on a busy free tier.
Conversational/knowledge answers (single AI call) keep Sasquatch's personality
and are untouched.
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- findCoaches now covers EVERY Lab craft (coach, filmer, editor, photographer,
builder, artist, graffiti, tattoo, trainer, musician, judge) via role param;
reads skills from `t` tags and strips self-tags.
- New findShops (kind 30017, t=skateshop) and findTours (kind 30823) tools.
- getCommunityStats counts all crafts individually + lab_total + shops + tours,
and filters spots to t=skatespot (drops foreign kind-30818 events).
- Wired new tools + expanded role vocabulary into the Sasquatch brain: system
prompt, intent parser, keyword detector, and both dispatch/format paths.
- Privacy preserved: the public CVM data layer never queries the encrypted
Stash (kind 30001) — verified no 30001/decrypt references. Personal 'my
stash' stays login+Pro gated and decrypts locally with the user's own key.
- Normalized useCVMClient.ts to clean UTF-8 (comment glyphs only; no code change)
Co-authored-by: shakespeare.diy <assistant@shakespeare.diy>
- useCVMClient: add findCrews (queries kind 35006 across data relays, filters by
city/country/vibe/search, returns CrewResult with member_count, vibes, url);
add crews to getCommunityStats + CrewResult type
- SasquatchGuide: register find_crews tool in system prompt + switch handler,
add crew keywords to needsLiveData, update app-facts (crews are searchable +
mention cross-linking and Zap-the-Crew)
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>
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>
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>
- 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>
