nodebench-mcp 2.70.0 → 3.0.1

package/rules/nodebench-pre-release-review.md

@@ -0,0 +1,24 @@
+# NodeBench: Pre-Release Review
+
+Run before any deploy, PR, or demo.
+
+## 7-Layer Review Stack
+
+1. **Build gate** — `npx tsc --noEmit` + `npx vite build` (0 errors)
+2. **Test gate** — `npm run test:run` (0 failures)
+3. **Visual sweep** — Visit every surface, check for console errors, layout shift, contrast
+4. **Agent panel** — Open agent panel, run demo conversation, close cleanly
+5. **Content freshness** — No stale dates, TODOs, "Coming soon", developer jargon in UI
+6. **A11y spot-check** — Tab through page, check focus rings, contrast, reduced motion
+7. **Bundle sanity** — No chunks >500KB, route chunks separate
+
+## Failure protocol
+
+- Layer 1-2 failures: BLOCK. Fix before any review.
+- Layer 3-7 failures: FIX INLINE with before/after evidence.
+
+## NodeBench tools
+
+- `run_quality_gate()` — automated checks
+- `capture_responsive_suite({ url })` — visual proof at all breakpoints
+- `site_map({ url })` — full surface coverage

package/rules/nodebench-qa-dogfood.md

@@ -0,0 +1,26 @@
+# NodeBench: QA Dogfood Protocol
+
+After ANY implementation change, run this checklist before declaring done.
+
+## Automated checks
+
+1. `npx tsc --noEmit` — must be 0 errors
+2. `npx vitest run` — must be 0 failures
+3. Start dev server and verify all surfaces render
+4. Check console for errors on every surface
+5. Test navigation: rail switches, CTA buttons, demo cards
+
+## NodeBench tool-assisted QA
+
+- `run_quality_gate()` — automated quality checks
+- `capture_responsive_suite({ url })` — screenshots at mobile/tablet/desktop
+- `run_visual_qa_suite()` — Gemini vision scoring against design principles
+- `site_map({ url })` — crawl and inspect all pages interactively
+- `diff_crawl({ url })` — before/after comparison
+
+## Never declare work complete without
+
+- Running the build (0 errors)
+- Running tests (0 failures)
+- Visual verification (screenshot or preview)
+- Console check (0 errors on all surfaces)

package/rules/nodebench-scenario-testing.md

@@ -0,0 +1,30 @@
+# NodeBench: Scenario-Based Testing
+
+Never write simple tests. Every test must be scenario-based.
+
+## Every test must answer 6 questions
+
+1. **Who** — Which user persona? (first-timer, power user, adversarial, mobile)
+2. **What** — What is the user trying to achieve? (goal, not function)
+3. **How** — Action sequence, timing, concurrency pattern
+4. **Scale** — 1 user, 10 concurrent, 100 sustained
+5. **Duration** — Short-running (burst) AND long-running (accumulation)
+6. **Failure modes** — Edge cases, race conditions, degraded inputs
+
+## Required categories per feature
+
+| Category | What it covers |
+|---|---|
+| Happy path | Baseline correctness |
+| Sad paths | Every error condition, boundary value |
+| Concurrent | Race conditions, dirty reads, double-submit |
+| Degraded | Slow network, auth expiry, partial failures |
+| Long-running | State after 100+ sessions, memory growth |
+| Adversarial | Injection, replay attacks, unexpected payloads |
+
+## Anti-patterns (banned)
+
+- Tests with no defined user persona
+- Tests that mock everything and test nothing real
+- Tests that pass at 1 user and are never run at 10+
+- Declaring "tested" after a single happy-path check

package/rules/nodebench-self-direction.md

@@ -0,0 +1,23 @@
+# NodeBench: Self-Direction
+
+Never wait for permission, confirmation, or next instructions. Direct yourself.
+
+## Protocol
+
+1. **Decide** — When a task completes, immediately identify the next highest-impact action
+2. **Act** — Start the next task. If blocked, pivot to a parallel workstream
+3. **Visual-first** — Verify work visually (screenshots, previews, e2e tests). Code grep is supplementary.
+4. **Momentum** — Batch related fixes. Ship small increments. Keep the build green.
+5. **Escalate only on ambiguity** — Only pause when direction is genuinely unclear
+
+## NodeBench tools for momentum
+
+- `discover_tools('next step')` — find what to do next
+- `get_workflow_chain('verify')` — step-by-step verification workflow
+- `get_system_pulse()` — quick health check before moving on
+
+## Anti-patterns
+
+- "What would you like me to do next?"
+- Listing findings without acting on them
+- Running code searches when screenshots would answer faster

package/rules/nodebench-self-judge-loop.md

@@ -0,0 +1,24 @@
+# NodeBench: Self-Judge Loop
+
+Never declare work complete without running it through a structured judge pipeline.
+
+## Protocol
+
+1. **Find the methods yourself** — Search for analytical frameworks used by top practitioners
+2. **Produce multiple versions** — At least 3 competing approaches with tradeoffs
+3. **Deconstruct best in class** — Why did the best products in adjacent spaces succeed?
+4. **Build structured judge pipelines** — Boolean-scored evaluation, loop until all pass
+5. **Audit the product, not just the artifact** — Time to dopamine, feedback loop speed, shareability
+
+## NodeBench tools for self-judging
+
+- `run_quality_gate()` — automated boolean checks
+- `run_visual_qa_suite()` — Gemini vision scoring
+- `compare_savings()` — measure improvement over time
+- `suggest_tests()` — generate test cases from crawl findings
+
+## Anti-patterns
+
+- Producing one version and asking "is this good?"
+- Evaluating with made-up criteria instead of industry frameworks
+- Waiting for the user to say "also check X"

package/scripts/install.sh

@@ -0,0 +1,215 @@
+#!/usr/bin/env bash
+# NodeBench MCP — One-liner installer
+# Usage: curl -sL https://nodebenchai.com/install.sh | bash
+#   or:  bash packages/mcp-local/scripts/install.sh
+
+set -euo pipefail
+
+# ── Colors ──────────────────────────────────────────────────────────────
+if [ -t 1 ]; then
+  B="\033[1m"
+  G="\033[32m"
+  Y="\033[33m"
+  C="\033[36m"
+  R="\033[31m"
+  X="\033[0m"
+else
+  B="" G="" Y="" C="" R="" X=""
+fi
+
+echo -e "${B}NodeBench MCP — Installer${X}"
+echo ""
+
+# ── Prerequisites ───────────────────────────────────────────────────────
+# Check Node.js
+if ! command -v node &>/dev/null; then
+  echo -e "${R}ERROR${X}: Node.js not found. Install Node.js >= 18: https://nodejs.org"
+  exit 1
+fi
+
+NODE_VERSION=$(node -v | sed 's/v//' | cut -d. -f1)
+if [ "$NODE_VERSION" -lt 18 ]; then
+  echo -e "${R}ERROR${X}: Node.js >= 18 required (found v$(node -v))"
+  exit 1
+fi
+echo -e "  ${G}OK${X}  Node.js $(node -v)"
+
+# Check npm
+if ! command -v npm &>/dev/null; then
+  echo -e "${R}ERROR${X}: npm not found"
+  exit 1
+fi
+echo -e "  ${G}OK${X}  npm $(npm -v)"
+
+# ── Detect user info ────────────────────────────────────────────────────
+USER_EMAIL=$(git config user.email 2>/dev/null || echo "")
+if [ -n "$USER_EMAIL" ]; then
+  echo -e "  ${G}OK${X}  Git email: ${C}${USER_EMAIL}${X}"
+fi
+
+# ── Install nodebench-mcp globally ──────────────────────────────────────
+echo ""
+echo -e "${B}Installing nodebench-mcp...${X}"
+npm install -g nodebench-mcp@latest 2>/dev/null || {
+  echo -e "${Y}WARN${X}: Global install failed, will use npx instead"
+}
+
+# ── Determine nodebench-mcp location ────────────────────────────────────
+NODEBENCH_PATH=""
+if command -v nodebench-mcp &>/dev/null; then
+  NODEBENCH_PATH=$(which nodebench-mcp)
+  echo -e "  ${G}OK${X}  Installed: ${NODEBENCH_PATH}"
+else
+  echo -e "  ${G}OK${X}  Will use: npx -y nodebench-mcp"
+fi
+
+# ── Find rules directory ────────────────────────────────────────────────
+RULES_SOURCE=""
+if [ -n "$NODEBENCH_PATH" ]; then
+  # Global install — rules are in the package
+  PKG_DIR=$(dirname "$(dirname "$NODEBENCH_PATH")")/lib/node_modules/nodebench-mcp
+  if [ -d "$PKG_DIR/rules" ]; then
+    RULES_SOURCE="$PKG_DIR/rules"
+  fi
+fi
+
+# Fallback: try npm root
+if [ -z "$RULES_SOURCE" ]; then
+  NPM_ROOT=$(npm root -g 2>/dev/null || echo "")
+  if [ -d "$NPM_ROOT/nodebench-mcp/rules" ]; then
+    RULES_SOURCE="$NPM_ROOT/nodebench-mcp/rules"
+  fi
+fi
+
+# Fallback: npx cache
+if [ -z "$RULES_SOURCE" ]; then
+  # Run npx once to ensure package is cached, then find it
+  npx -y nodebench-mcp --help >/dev/null 2>&1 || true
+  NPX_CACHE=$(find "$HOME/.npm/_npx" -name "nodebench-mcp" -type d 2>/dev/null | head -1)
+  if [ -n "$NPX_CACHE" ] && [ -d "$NPX_CACHE/rules" ]; then
+    RULES_SOURCE="$NPX_CACHE/rules"
+  fi
+fi
+
+# ── Install rules to ~/.claude/rules/ ───────────────────────────────────
+echo ""
+CLAUDE_RULES_DIR="$HOME/.claude/rules"
+mkdir -p "$CLAUDE_RULES_DIR"
+
+RULES_INSTALLED=0
+if [ -n "$RULES_SOURCE" ] && [ -d "$RULES_SOURCE" ]; then
+  echo -e "${B}Installing rules to ${CLAUDE_RULES_DIR}/${X}"
+  for rule in "$RULES_SOURCE"/nodebench-*.md; do
+    [ -f "$rule" ] || continue
+    BASENAME=$(basename "$rule")
+    TARGET="$CLAUDE_RULES_DIR/$BASENAME"
+    # Don't overwrite if user's version is newer
+    if [ -f "$TARGET" ] && [ "$TARGET" -nt "$rule" ]; then
+      echo -e "  ${Y}SKIP${X} $BASENAME (user version is newer)"
+    else
+      cp "$rule" "$TARGET"
+      echo -e "  ${G}OK${X}   $BASENAME"
+      RULES_INSTALLED=$((RULES_INSTALLED + 1))
+    fi
+  done
+  echo -e "  Installed ${G}${RULES_INSTALLED}${X} rules"
+else
+  echo -e "${Y}WARN${X}: Could not find rules directory. Run ${C}nodebench-mcp --sync-configs${X} after install to copy rules."
+fi
+
+# ── Write .mcp.json ─────────────────────────────────────────────────────
+echo ""
+MCP_CONFIG=".mcp.json"
+
+# Collect preset preference
+PRESET="starter"
+echo -e "${B}Select a preset:${X}"
+echo "  1) starter    — 15 tools (decision intelligence core)"
+echo "  2) founder    — 40 tools (weekly resets, delegation, entities)"
+echo "  3) researcher — 32 tools (web search, LLM, RSS, email)"
+echo "  4) web_dev    — 150 tools (vision, SEO, git, UI/UX)"
+echo "  5) full       — 338 tools (everything)"
+echo ""
+read -r -p "Choose [1-5, default=1]: " CHOICE </dev/tty 2>/dev/null || CHOICE="1"
+case "${CHOICE:-1}" in
+  2) PRESET="founder" ;;
+  3) PRESET="researcher" ;;
+  4) PRESET="web_dev" ;;
+  5) PRESET="full" ;;
+  *) PRESET="starter" ;;
+esac
+
+# Build MCP config
+if [ -n "$NODEBENCH_PATH" ]; then
+  MCP_CMD="nodebench-mcp"
+  MCP_ARGS="[\"--preset\", \"$PRESET\"]"
+else
+  MCP_CMD="npx"
+  MCP_ARGS="[\"-y\", \"nodebench-mcp\", \"--preset\", \"$PRESET\"]"
+fi
+
+# Collect env vars
+ENV_BLOCK=""
+if [ -n "${GEMINI_API_KEY:-}" ]; then
+  ENV_BLOCK="\"GEMINI_API_KEY\": \"$GEMINI_API_KEY\""
+fi
+if [ -n "${OPENAI_API_KEY:-}" ]; then
+  [ -n "$ENV_BLOCK" ] && ENV_BLOCK="$ENV_BLOCK, "
+  ENV_BLOCK="${ENV_BLOCK}\"OPENAI_API_KEY\": \"$OPENAI_API_KEY\""
+fi
+if [ -n "${GITHUB_TOKEN:-}" ]; then
+  [ -n "$ENV_BLOCK" ] && ENV_BLOCK="$ENV_BLOCK, "
+  ENV_BLOCK="${ENV_BLOCK}\"GITHUB_TOKEN\": \"$GITHUB_TOKEN\""
+fi
+
+ENV_JSON="{}"
+if [ -n "$ENV_BLOCK" ]; then
+  ENV_JSON="{$ENV_BLOCK}"
+fi
+
+# Write or merge .mcp.json
+if [ -f "$MCP_CONFIG" ]; then
+  echo -e "${Y}NOTE${X}: $MCP_CONFIG already exists. Adding nodebench entry."
+fi
+
+cat > "$MCP_CONFIG" <<MCPEOF
+{
+  "mcpServers": {
+    "nodebench": {
+      "command": "$MCP_CMD",
+      "args": $MCP_ARGS,
+      "env": $ENV_JSON
+    }
+  }
+}
+MCPEOF
+echo -e "  ${G}OK${X}  Written: ${C}${MCP_CONFIG}${X} (preset: $PRESET)"
+
+# ── Add .mcp.json to .gitignore ─────────────────────────────────────────
+if [ -f ".gitignore" ]; then
+  if ! grep -qF ".mcp.json" .gitignore; then
+    echo "" >> .gitignore
+    echo "# NodeBench MCP config (contains API keys)" >> .gitignore
+    echo ".mcp.json" >> .gitignore
+    echo -e "  ${G}OK${X}  Added .mcp.json to .gitignore"
+  fi
+fi
+
+# ── Run health check ────────────────────────────────────────────────────
+echo ""
+echo -e "${B}Running health check...${X}"
+npx -y nodebench-mcp --health 2>/dev/null || echo -e "${Y}WARN${X}: Health check skipped (will run on first use)"
+
+# ── Summary ─────────────────────────────────────────────────────────────
+echo ""
+echo -e "${B}${G}NodeBench MCP installed successfully!${X}"
+echo ""
+echo -e "${B}Try these first:${X}"
+echo -e "  ${C}discover_tools('analyze a company')${X}     — find relevant tools"
+echo -e "  ${C}load_toolset('founder')${X}                  — activate founder tools"
+echo -e "  ${C}get_workflow_chain('weekly_reset')${X}        — get guided workflow"
+echo -e "  ${C}site_map({ url: 'https://yoursite.com' })${X} — crawl & inspect"
+echo ""
+echo -e "  Run ${C}nodebench-mcp --health${X} anytime to check system status."
+echo -e "  Run ${C}nodebench-mcp --sync-configs${X} to write configs to Claude/Cursor/Windsurf."
+echo ""

package/dist/__tests__/analytics.test.d.ts

@@ -1,11 +0,0 @@
-/**
- * Comprehensive tests for the analytics subsystem.
- *
- * Tests cover:
- *   - schema.ts:          DB init, singleton, CRUD, cache TTL, retention cleanup
- *   - projectDetector.ts: project type detection, glob matching, edge cases
- *   - usageStats.ts:      aggregation queries, trends, failing tools, formatted display
- *   - smartPreset.ts:     signal computation, scoring model, formatted output
- *   - toolTracker.ts:     AnalyticsTracker singleton, record(), session stats, close
- */
-export {};