feat: NIP-50 search, NIP-10 threads, NIP-17 DMs, Sprout DM discovery

[tlongwell-block]

Summary

Make Sprout usable by standard Nostr clients. Four independently-shippable features, all reusing existing infrastructure — no new databases, no new services, no new tables. Plus iterative review-driven correctness hardening of the historical REQ and search paths.

22 commits · 13 files · +1826/−81 · 15 E2E tests · 93/93 E2E pass · 740+ unit tests pass

Nostr Client (WS/HTTP)                Sprout Relay / Existing Infra
────────────────────────────────────────────────────────────────────────────
REQ {search:"foo"}            ─────►  REQ handler ─────► Typesense
                                         │                 (pre-existing)
                                         └──── batch DB fetch by event IDs

kind:9 + NIP-10 e-tags       ─────►  EVENT handler ───► insert_event_with_thread_metadata
                                                            (pre-existing DB path)

kind:1059 + #p               ─────►  EVENT / REQ handlers
                                         │
                                         ├──── stored in events table (pre-existing)
                                         ├──── gated by #p via event_mentions join
                                         └──── excluded from search/workflows

POST /api/dms + add member   ─────►  existing DM REST paths
                                         │
                                         └──── emit NIP-29 discovery + 44100 membership events

Review process: iterative Codex crossfire: 5 → 7 → 8 → 7 → 7 → 6 → 8 → 7 → 8 → 7 → 8 → 7 → 8 → 8 → 9. Most commits after the initial feature are direct responses to review findings.

---

What Changed

1. Direct-path NIP support

NIP-50 Search (one-shot, Typesense-backed)

• Search filters handled as one-shot — not registered as persistent subscriptions (avoids filters_match bug where persistent search subs match all future events)
• Hits fetched from Typesense, resolved to full events from MySQL, delivered in relevance order
• Pagination loop (up to 10 pages × 100 per page) ensures post-filtering doesn't silently reduce result count
• authors, since, until, kinds, and per-filter #h pushed into Typesense filter_by
• #h values intersected with accessible channels; invalid/inaccessible #h matches nothing (not all)
• limit == 0 handled explicitly
• Search intercept runs before #p gating so wildcard-kind searches aren't spuriously blocked

Files: handlers/req.rs, sprout-db/event.rs (batch fetch), sprout-db/lib.rs

NIP-10 Threads (pre-storage ancestry resolution)

• Scans e tags for NIP-10 root/reply markers before storage
• Uses existing atomic insert_event_with_thread_metadata — prevents documented race condition
• Root-only e tags treated as non-threaded context (not replies)
• Client-supplied root validated against server-resolved ancestry; rejected if they diverge
• Legacy fallback: handles direct-reply form (["e",<root>,"","reply"]) when parent has no thread_metadata
• Unknown/cross-channel parents rejected (parity with REST)
• Thread depth capped at 100

Files: handlers/event.rs

NIP-17 DMs (opaque gift-wrap storage)

• kind:1059 exempt from pubkey match (ephemeral signing keys by design)
• Forced to channel_id = None — client h tags ignored (prevents channel-scoped bypass of #p gating)
• Excluded from Typesense indexing and workflow evaluation
• P_GATED_KINDS consolidates #p enforcement for kind:44100, 44101, and 1059

Files: handlers/event.rs, handlers/req.rs

Sprout DM Discovery

• open_dm_handler and add_dm_member_handler emit NIP-29 discovery events on new DM channels
• kind:39000 metadata includes hidden tag for DMs (client hint, not security boundary)
• Each participant receives kind:44100 membership notification

Files: api/dms.rs, handlers/side_effects.rs

---

2. Review-driven correctness fixes

Typesense multi_search

Switched from GET to POST /multi_search. GET has a 4000-char query string limit; with 400+ accessible channels the filter_by exceeds this. Affects both NIP-50 and the existing REST /api/search.

Files: sprout-search/query.rs

Historical REQ hardening

• Per-filter #h pushdown — extracts channel from each filter's #h tag instead of using the subscription-level channel_id (prevents under-fetch when filters target different channels)
• Per-filter post-matching instead of OR across all filters (was causing under-fetch)
• Dedup after acceptance — event that fails filter A remains eligible for filter B (NIP-01 OR semantics)
• #p pushdown via event_mentions JOIN — gift-wrap/membership queries no longer miss results after 500-row LIMIT
• Single-author SQL pushdown — pubkey pushed into WHERE clause for single-author filters
• kinds:[] returns zero rows — empty-kinds sentinel now honored by DB layer (was matching all kinds)

Files: handlers/req.rs, sprout-db/event.rs

---

3. Docs and tests

• NOSTR.md updated: threads/search/DMs moved from "What Doesn't Work" to "What Works" with nak examples
• NIP-11 now advertises [1, 10, 11, 17, 25, 29, 42, 50]
• 15 E2E tests covering all 4 features plus regression tests for:
  • Thread reply hidden from top-level messages
  • Gift wrap not indexed in Typesense (verified via direct Typesense query)
  • Search relevance ordering (first-position assertion)
  • Historical dedup preserves OR semantics across filters
  • kinds:[] returns zero events
• Existing E2E fixes: max_subscriptions assertion (1024), search indexing delay (2s)

---

Testing

Suite	Result
e2e_nostr_interop (new)	15/15 ✅
e2e_relay	27/27 ✅
e2e_rest_api	51/51 ✅
e2e_mcp	14/14 ✅
Unit tests (workspace)	740+ pass, 0 fail

Live testing

• nak CLI: kind:9 send/recv, NIP-50 search, NIP-10 thread replies, group discovery
• ACP round-trip: user → relay → ACP harness → goose agent → MCP → relay → reply visible (echo: nak-e2e-...)

---

What we're NOT doing

• NIP-04 / NIP-44 DM support — NIP-17 only
• kind:10050 relay-list — requires generic replaceable-event semantics
• Proxied NIP-50 — direct path only
• Live/persistent search subscriptions — one-shot by design
• Bridging Sprout DMs ↔ NIP-17 DMs — separate systems