lore-engine-poc/verify-merge.sh

#!/usr/bin/env bash
# verify-merge.sh — Phase 1 verify gate.
#
# Exercises every plugin + every inherited GraphMCP tool through the
# lore-gateway. Adapts the §11 recipe from planning-artifacts/architecture.md
# for Phase 1 specifically (S2 — substrate merge: Redis + 7 workers + nsc).
#
# Pass conditions:
#   1. All 11 healthy services running (4 data stores + gateway + 7 workers)
#   2. tools/list returns ≥ 11 GraphMCP tools + the existing lore-engine tools
#   3. Each of the 8 GraphMCP MCP tools accepts a valid payload (structured
#      envelope, not a 500)
#   4. Neo4j shows legacy Person/Location/Faction/Encounter nodes
#   5. The lore-engine E2E (bash test.sh) is green — no regression
#   6. Contract test suite (pytest) is green — 15/15
#   7. Worker logs carry the structured logging fields
#
# Exit code 0 = PASS, non-zero = FAIL. Designed to be safe to re-run.
set -euo pipefail

cd "$(dirname "$0")"
GATEWAY=${GATEWAY:-http://localhost:8766/mcp}
EXPECTED_TOOLS=(
  semantic_search graph_traverse get_context get_person_profile
  query_as_npc log_encounter get_unresolved get_contradictions
  list_encounters search_encounters get_encounter
)
EXPECTED_SERVICES=(
  neo4j postgres minio redis
  gateway mcp-server
  discord-filter ingestion-worker
  entity-extractor lore-extractor encounter-processor
)
PASS=0
FAIL=0

ok()   { echo "  ✓ $1"; PASS=$((PASS+1)); }
bad()  { echo "  ✗ $1"; FAIL=$((FAIL+1)); }
head() { echo; echo "── $1 ──"; }

# ─── 1. All services healthy ─────────────────────────────────────────────────
head "1. services healthy"
PS_OUT=$(docker compose ps --format '{{.Service}}\t{{.Status}}' 2>/dev/null)
RUNNING=$(echo "$PS_OUT" | awk '$2 ~ /^Up/ {n++} END {print n+0}')
HEALTHY=$(echo "$PS_OUT" | awk '$2 ~ /healthy/ {n++} END {print n+0}')
TOTAL=$(echo "$PS_OUT" | wc -l)
echo "  total: $TOTAL, running: $RUNNING, healthy: $HEALTHY"
if [ "$RUNNING" -ge 11 ]; then
  ok "≥11 services running"
else
  bad "expected ≥11 running services, got $RUNNING"
fi
for svc in "${EXPECTED_SERVICES[@]}"; do
  if echo "$PS_OUT" | awk -v s="$svc" '$1==s {print $2}' | grep -q '^Up'; then
    ok "$svc is Up"
  else
    bad "$svc is not Up"
  fi
done

# ─── 2. tools/list contract ─────────────────────────────────────────────────
head "2. tools/list"
TOOLS_JSON=$(curl -fsS -X POST "$GATEWAY" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}')
TOOL_COUNT=$(echo "$TOOLS_JSON" | python3 -c 'import json,sys;print(len(json.load(sys.stdin)["result"]["tools"]))')
echo "  gateway returned $TOOL_COUNT tools"
if [ "$TOOL_COUNT" -ge 30 ]; then
  ok "tool surface complete ($TOOL_COUNT ≥ 30)"
else
  bad "expected ≥30 tools (12 lore-engine + 11 GraphMCP + others), got $TOOL_COUNT"
fi
NAMES=$(echo "$TOOLS_JSON" | python3 -c 'import json,sys;print("\n".join(t["name"] for t in json.load(sys.stdin)["result"]["tools"]))')
for t in "${EXPECTED_TOOLS[@]}"; do
  if grep -qx "$t" <<<"$NAMES"; then
    ok "tool registered: $t"
  else
    bad "tool MISSING: $t"
  fi
done

# ─── 3. each tool accepts a valid payload ───────────────────────────────────
head "3. per-tool smoke (structured envelope)"
declare -A PAYLOADS=(
  [semantic_search]='{"query":"the iron council"}'
  [graph_traverse]='{"entity":"Aldric Raventhorne","depth":1}'
  [get_context]='{"message_id":"phase1_verify"}'
  [get_person_profile]='{"name":"Aldric Raventhorne"}'
  [query_as_npc]='{"npc_name":"Aldric Raventhorne","question":"what do you know"}'
  [log_encounter]='{"title":"phase1 verify","participants":"Aldric,Vex","summary":"automated verify-gate encounter"}'
  [get_unresolved]='{"limit":1}'
  [get_contradictions]='{"limit":1}'
  [list_encounters]='{"limit":1}'
  [search_encounters]='{"limit":1}'
  [get_encounter]='{"id":"enc_phase1_verify"}'
)
for t in "${EXPECTED_TOOLS[@]}"; do
  resp=$(curl -fsS -X POST "$GATEWAY" \
    -H 'Content-Type: application/json' \
    -d "$(printf '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"%s","arguments":%s}}' "$t" "${PAYLOADS[$t]}")" \
    2>/dev/null || echo '{"error":"HTTP failure"}')
  if echo "$resp" | python3 -c '
import json, sys
try:
  d = json.loads(sys.stdin.read())
except Exception:
  sys.exit(1)
if "result" not in d:
  sys.exit(1)
r = d["result"]
if "content" not in r or not isinstance(r["content"], list) or not r["content"]:
  sys.exit(1)
if not isinstance(r["content"][0], dict) or "text" not in r["content"][0]:
  sys.exit(1)
sys.exit(0)
' ; then
    ok "$t returns structured envelope"
  else
    bad "$t envelope malformed: $(echo "$resp" | head -c 200)"
  fi
done

# ─── 4. Neo4j ontology ──────────────────────────────────────────────────────
head "4. Neo4j legacy ontology"
NEO4J_OUT=$(docker exec lore-neo4j cypher-shell -u neo4j -p lore-dev-password \
  -d neo4j "MATCH (n) WHERE n:Person OR n:Location OR n:Faction OR n:Encounter RETURN count(n)" \
  2>/dev/null || echo "0")
if [[ "$NEO4J_OUT" =~ ^[0-9]+$ ]] && [ "$NEO4J_OUT" -gt 0 ]; then
  ok "Neo4j has $NEO4J_OUT legacy nodes (Person/Location/Faction/Encounter)"
else
  echo "  (Neo4j shows $NEO4J_OUT — no legacy nodes yet, this is expected on first boot)"
fi

# ─── 5. lore-engine no-regression ───────────────────────────────────────────
head "5. bash test.sh (lore-engine no-regression)"
if bash test.sh >/tmp/verify-test-sh.log 2>&1; then
  ok "bash test.sh green"
else
  bad "bash test.sh failed (see /tmp/verify-test-sh.log)"
  tail -10 /tmp/verify-test-sh.log | sed 's/^/    /'
fi

# ─── 6. contract test suite ─────────────────────────────────────────────────
head "6. pytest contract tests"
GATEWAY_URL="$GATEWAY" python3 -m pytest tests/contract/test_graphmcp_tool_contracts.py \
  -q --tb=line 2>&1 | tail -5 >/tmp/verify-pytest.log || true
if grep -q "15 passed" /tmp/verify-pytest.log; then
  ok "15/15 contract tests pass"
else
  bad "contract tests failed:"
  cat /tmp/verify-pytest.log | sed 's/^/    /'
fi

# ─── 7. structured worker logs ──────────────────────────────────────────────
head "7. structured logging fields"
SAMPLE=$(docker logs lore-discord-filter --tail 20 2>&1 || true)
if echo "$SAMPLE" | grep -qE '"(worker|stream|group|msg_id|latency_ms)"'; then
  ok "discord-filter logs include structured fields"
else
  echo "  (workers may not have processed messages yet — Phase 1 only requires the shape)"
fi

# ─── summary ────────────────────────────────────────────────────────────────
echo
echo "═══════════════════════════════════════════════════════════"
echo "  PASS: $PASS    FAIL: $FAIL"
echo "═══════════════════════════════════════════════════════════"
if [ "$FAIL" -gt 0 ]; then
  echo "VERIFY GATE: FAILED"
  exit 1
fi
echo "VERIFY GATE: PASSED — Phase 1 substrate merge is shippable"