@jonit-dev/night-watch-cli 1.7.9 → 1.7.10

package/scripts/night-watch-cron.sh

@@ -290,9 +290,14 @@ fi
 BACKOFF_BASE=300  # 5 minutes in seconds
 EXIT_CODE=0
 ATTEMPT=0
+RATE_LIMIT_FALLBACK_TRIGGERED=0
 
 while [ "${ATTEMPT}" -lt "${MAX_RETRIES}" ]; do
   EXIT_CODE=0
+  # Capture log position before this attempt so check_rate_limited only
+  # scans lines written by the current invocation (not leftover 429s from
+  # previous runs that would cause false-positive rate-limit retries).
+  LOG_LINE_BEFORE=$(wc -l < "${LOG_FILE}" 2>/dev/null || echo 0)
 
   case "${PROVIDER_CMD}" in
     claude)
@@ -331,8 +336,14 @@ while [ "${ATTEMPT}" -lt "${MAX_RETRIES}" ]; do
     break
   fi
 
-  # Check if this was a rate limit (429) error
-  if check_rate_limited "${LOG_FILE}"; then
+  # Check if this was a rate limit (429) error (only in lines from this attempt)
+  if check_rate_limited "${LOG_FILE}" "${LOG_LINE_BEFORE}"; then
+    # If fallback is enabled, skip proxy retries and switch to native Claude immediately
+    if [ "${NW_FALLBACK_ON_RATE_LIMIT:-}" = "true" ] && [ "${PROVIDER_CMD}" = "claude" ]; then
+      log "RATE-LIMITED: Proxy quota exhausted — triggering native Claude fallback"
+      RATE_LIMIT_FALLBACK_TRIGGERED=1
+      break
+    fi
     ATTEMPT=$((ATTEMPT + 1))
     if [ "${ATTEMPT}" -ge "${MAX_RETRIES}" ]; then
       log "RATE-LIMITED: All ${MAX_RETRIES} attempts exhausted for ${ELIGIBLE_PRD}"
@@ -349,6 +360,36 @@ while [ "${ATTEMPT}" -lt "${MAX_RETRIES}" ]; do
   fi
 done
 
+# ── Native Claude fallback ────────────────────────────────────────────────────
+# When the proxy returns 429 and fallbackOnRateLimit is enabled, re-run the
+# same prompt with native Claude (OAuth), bypassing the proxy entirely.
+if [ "${RATE_LIMIT_FALLBACK_TRIGGERED}" = "1" ]; then
+  FALLBACK_MODEL="${NW_CLAUDE_MODEL_ID:-claude-sonnet-4-6}"
+  log "RATE-LIMIT-FALLBACK: Running native Claude (${FALLBACK_MODEL})"
+
+  # Send immediate Telegram warning (fire-and-forget)
+  send_rate_limit_fallback_warning "${FALLBACK_MODEL}" "$(basename "${PROJECT_DIR}")"
+
+  LOG_LINE_BEFORE=$(wc -l < "${LOG_FILE}" 2>/dev/null || echo 0)
+
+  if (
+    cd "${WORKTREE_DIR}" && \
+      unset ANTHROPIC_BASE_URL ANTHROPIC_API_KEY ANTHROPIC_AUTH_TOKEN \
+            ANTHROPIC_DEFAULT_SONNET_MODEL ANTHROPIC_DEFAULT_OPUS_MODEL && \
+      timeout "${MAX_RUNTIME}" \
+        claude -p "${PROMPT}" \
+          --dangerously-skip-permissions \
+          --model "${FALLBACK_MODEL}" \
+          >> "${LOG_FILE}" 2>&1
+  ); then
+    EXIT_CODE=0
+  else
+    EXIT_CODE=$?
+  fi
+
+  log "RATE-LIMIT-FALLBACK: Native Claude exited with code ${EXIT_CODE}"
+fi
+
 if [ ${EXIT_CODE} -eq 0 ]; then
   OPEN_PR_COUNT=$(count_prs_for_branch open "${BRANCH_NAME}")
   if [ "${OPEN_PR_COUNT}" -gt 0 ]; then

package/scripts/night-watch-helpers.sh

@@ -406,11 +406,34 @@ mark_prd_done() {
 
 # ── Rate limit detection ────────────────────────────────────────────────────
 
-# Check if the last N lines of the log contain a 429 rate limit error.
+# Check if the log contains a 429 rate limit error since a given line number.
+# Usage: check_rate_limited <log_file> [start_line]
+# When start_line is provided, only lines after that position are checked,
+# preventing false positives from 429 errors in previous runs.
 # Returns 0 if rate limited, 1 otherwise.
 check_rate_limited() {
   local log_file="${1:?log_file required}"
-  tail -20 "${log_file}" 2>/dev/null | grep -q "429"
+  local start_line="${2:-0}"
+  if [ "${start_line}" -gt 0 ] 2>/dev/null; then
+    tail -n "+$((start_line + 1))" "${log_file}" 2>/dev/null | grep -q "429"
+  else
+    tail -20 "${log_file}" 2>/dev/null | grep -q "429"
+  fi
+}
+
+# Send an immediate Telegram warning when the rate-limit fallback is triggered.
+# Uses NW_TELEGRAM_BOT_TOKEN and NW_TELEGRAM_CHAT_ID exported by the CLI runner.
+# Falls back silently when credentials are absent.
+# Usage: send_rate_limit_fallback_warning <model> <project_name>
+send_rate_limit_fallback_warning() {
+  local model="${1:-native Claude}"
+  local project_name="${2:-unknown}"
+  if [ -z "${NW_TELEGRAM_BOT_TOKEN:-}" ] || [ -z "${NW_TELEGRAM_CHAT_ID:-}" ]; then
+    return 0
+  fi
+  local msg="⚠️ Rate Limit Fallback%0A%0AProject: ${project_name}%0AProxy quota exhausted — falling back to native Claude (${model})"
+  curl -s -X POST "https://api.telegram.org/bot${NW_TELEGRAM_BOT_TOKEN}/sendMessage" \
+    -d "chat_id=${NW_TELEGRAM_CHAT_ID}&text=${msg}" > /dev/null 2>&1 || true
 }
 
 # ── Board mode issue discovery ────────────────────────────────────────────────

package/scripts/night-watch-pr-reviewer-cron.sh

@@ -10,6 +10,8 @@ set -euo pipefail
 #   NW_REVIEWER_MAX_RUNTIME=3600 - Maximum runtime in seconds (1 hour)
 #   NW_PROVIDER_CMD=claude       - AI provider CLI to use (claude, codex, etc.)
 #   NW_DRY_RUN=0                 - Set to 1 for dry-run mode (prints diagnostics only)
+#   NW_AUTO_MERGE=0              - Set to 1 to enable auto-merge
+#   NW_AUTO_MERGE_METHOD=squash  - Merge method: squash, merge, or rebase
 
 PROJECT_DIR="${1:?Usage: $0 /path/to/project}"
 PROJECT_NAME=$(basename "${PROJECT_DIR}")
@@ -20,6 +22,8 @@ MAX_LOG_SIZE="524288"  # 512 KB
 PROVIDER_CMD="${NW_PROVIDER_CMD:-claude}"
 MIN_REVIEW_SCORE="${NW_MIN_REVIEW_SCORE:-80}"
 BRANCH_PATTERNS_RAW="${NW_BRANCH_PATTERNS:-feat/,night-watch/}"
+AUTO_MERGE="${NW_AUTO_MERGE:-0}"
+AUTO_MERGE_METHOD="${NW_AUTO_MERGE_METHOD:-squash}"
 
 # Ensure NVM / Node / Claude are on PATH
 export NVM_DIR="${HOME}/.nvm"
@@ -156,6 +160,10 @@ if [ "${NW_DRY_RUN:-0}" = "1" ]; then
   echo "Provider: ${PROVIDER_CMD}"
   echo "Branch Patterns: ${BRANCH_PATTERNS_RAW}"
   echo "Min Review Score: ${MIN_REVIEW_SCORE}"
+  echo "Auto-merge: ${AUTO_MERGE}"
+  if [ "${AUTO_MERGE}" = "1" ]; then
+    echo "Auto-merge Method: ${AUTO_MERGE_METHOD}"
+  fi
   echo "Open PRs needing work:${PRS_NEEDING_WORK}"
   echo "Default Branch: ${DEFAULT_BRANCH}"
   echo "Review Worktree: ${REVIEW_WORKTREE_DIR}"
@@ -204,6 +212,68 @@ esac
 
 cleanup_worktrees "${PROJECT_DIR}"
 
+# ── Auto-merge eligible PRs ─────────────────────────────────────────────────────
+# After the reviewer completes, check for PRs that are merge-ready and queue them
+# for auto-merge if enabled. Uses gh pr merge --auto to respect GitHub branch protection.
+AUTO_MERGED_PRS=""
+AUTO_MERGE_FAILED_PRS=""
+
+if [ "${AUTO_MERGE}" = "1" ] && [ ${EXIT_CODE} -eq 0 ]; then
+  log "AUTO-MERGE: Checking for merge-ready PRs..."
+
+  while IFS=$'\t' read -r pr_number pr_branch; do
+    if [ -z "${pr_number}" ] || [ -z "${pr_branch}" ]; then
+      continue
+    fi
+
+    # Only process PRs matching branch patterns
+    if ! printf '%s\n' "${pr_branch}" | grep -Eq "${BRANCH_REGEX}"; then
+      continue
+    fi
+
+    # Check CI status - must have no failures
+    FAILED_CHECKS=$(gh pr checks "${pr_number}" 2>/dev/null | grep -ci 'fail' || true)
+    if [ "${FAILED_CHECKS}" -gt 0 ]; then
+      continue
+    fi
+
+    # Check review score - must have score >= threshold
+    ALL_COMMENTS=$(
+      {
+        gh pr view "${pr_number}" --json comments --jq '.comments[].body' 2>/dev/null || true
+        if [ -n "${REPO}" ]; then
+          gh api "repos/${REPO}/issues/${pr_number}/comments" --jq '.[].body' 2>/dev/null || true
+        fi
+      } | sort -u
+    )
+    LATEST_SCORE=$(echo "${ALL_COMMENTS}" \
+      | grep -oP 'Overall Score:\*?\*?\s*(\d+)/100' \
+      | tail -1 \
+      | grep -oP '\d+(?=/100)' || echo "")
+
+    # Skip PRs without a score
+    if [ -z "${LATEST_SCORE}" ]; then
+      continue
+    fi
+
+    # Skip PRs with score below threshold
+    if [ "${LATEST_SCORE}" -lt "${MIN_REVIEW_SCORE}" ]; then
+      continue
+    fi
+
+    # PR is merge-ready - queue for auto-merge
+    log "AUTO-MERGE: PR #${pr_number} (${pr_branch}) — score ${LATEST_SCORE}/100, CI passing"
+
+    if gh pr merge "${pr_number}" --"${AUTO_MERGE_METHOD}" --auto --delete-branch 2>>"${LOG_FILE}"; then
+      log "AUTO-MERGE: Successfully queued merge for PR #${pr_number}"
+      AUTO_MERGED_PRS="${AUTO_MERGED_PRS} #${pr_number}"
+    else
+      log "WARN: Auto-merge failed for PR #${pr_number}"
+      AUTO_MERGE_FAILED_PRS="${AUTO_MERGE_FAILED_PRS} #${pr_number}"
+    fi
+  done < <(gh pr list --state open --json number,headRefName --jq '.[] | [.number, .headRefName] | @tsv' 2>/dev/null || true)
+fi
+
 if [ ${EXIT_CODE} -eq 0 ]; then
   log "DONE: PR reviewer completed successfully"
   emit_result "success_reviewed" "prs=${PRS_NEEDING_WORK_CSV}"

package/scripts/night-watch-qa-cron.sh

@@ -0,0 +1,269 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Night Watch QA Cron Runner (project-agnostic)
+# Usage: night-watch-qa-cron.sh /path/to/project
+#
+# NOTE: This script expects environment variables to be set by the caller.
+# The Node.js CLI will inject config values via environment variables.
+# Required env vars (with defaults shown):
+#   NW_QA_MAX_RUNTIME=3600            - Maximum runtime in seconds (1 hour)
+#   NW_PROVIDER_CMD=claude            - AI provider CLI to use (claude, codex, etc.)
+#   NW_BRANCH_PATTERNS=feat/,night-watch/ - Comma-separated branch prefixes to match
+#   NW_QA_SKIP_LABEL=skip-qa          - Label to skip QA on a PR
+#   NW_QA_ARTIFACTS=both              - Artifact mode (both, tests, report)
+#   NW_QA_AUTO_INSTALL_PLAYWRIGHT=1   - Auto-install Playwright browsers
+#   NW_DRY_RUN=0                      - Set to 1 for dry-run mode (prints diagnostics only)
+
+PROJECT_DIR="${1:?Usage: $0 /path/to/project}"
+PROJECT_NAME=$(basename "${PROJECT_DIR}")
+LOG_DIR="${PROJECT_DIR}/logs"
+LOG_FILE="${LOG_DIR}/night-watch-qa.log"
+MAX_RUNTIME="${NW_QA_MAX_RUNTIME:-3600}"  # 1 hour
+MAX_LOG_SIZE="524288"  # 512 KB
+PROVIDER_CMD="${NW_PROVIDER_CMD:-claude}"
+BRANCH_PATTERNS_RAW="${NW_BRANCH_PATTERNS:-feat/,night-watch/}"
+SKIP_LABEL="${NW_QA_SKIP_LABEL:-skip-qa}"
+QA_ARTIFACTS="${NW_QA_ARTIFACTS:-both}"
+QA_AUTO_INSTALL_PLAYWRIGHT="${NW_QA_AUTO_INSTALL_PLAYWRIGHT:-1}"
+
+# Ensure NVM / Node / Claude are on PATH
+export NVM_DIR="${HOME}/.nvm"
+[ -s "${NVM_DIR}/nvm.sh" ] && . "${NVM_DIR}/nvm.sh"
+
+# NOTE: Environment variables should be set by the caller (Node.js CLI).
+# The .env.night-watch sourcing has been removed - config is now injected via env vars.
+
+mkdir -p "${LOG_DIR}"
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=night-watch-helpers.sh
+source "${SCRIPT_DIR}/night-watch-helpers.sh"
+PROJECT_RUNTIME_KEY=$(project_runtime_key "${PROJECT_DIR}")
+# NOTE: Lock file path must match qaLockPath() in src/utils/status-data.ts
+LOCK_FILE="/tmp/night-watch-qa-${PROJECT_RUNTIME_KEY}.lock"
+
+emit_result() {
+  local status="${1:?status required}"
+  local details="${2:-}"
+  if [ -n "${details}" ]; then
+    echo "NIGHT_WATCH_RESULT:${status}|${details}"
+  else
+    echo "NIGHT_WATCH_RESULT:${status}"
+  fi
+}
+
+# Validate provider
+if ! validate_provider "${PROVIDER_CMD}"; then
+  echo "ERROR: Unknown provider: ${PROVIDER_CMD}" >&2
+  exit 1
+fi
+
+rotate_log
+
+if ! acquire_lock "${LOCK_FILE}"; then
+  emit_result "skip_locked"
+  exit 0
+fi
+
+cd "${PROJECT_DIR}"
+
+# Convert comma-separated branch prefixes into a regex that matches branch starts.
+BRANCH_REGEX=""
+IFS=',' read -r -a BRANCH_PATTERNS <<< "${BRANCH_PATTERNS_RAW}"
+for pattern in "${BRANCH_PATTERNS[@]}"; do
+  trimmed_pattern=$(printf '%s' "${pattern}" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//')
+  if [ -n "${trimmed_pattern}" ]; then
+    BRANCH_REGEX="${BRANCH_REGEX}${BRANCH_REGEX:+|}^${trimmed_pattern}"
+  fi
+done
+
+if [ -z "${BRANCH_REGEX}" ]; then
+  BRANCH_REGEX='^(feat/|night-watch/)'
+fi
+
+# List open PRs with their details for filtering
+PR_JSON=$(gh pr list --state open --json number,headRefName,title,labels 2>/dev/null || echo "[]")
+
+# Count PRs matching branch patterns
+OPEN_PRS=$(
+  echo "${PR_JSON}" \
+    | jq -r '.[].headRefName' 2>/dev/null \
+    | { grep -E "${BRANCH_REGEX}" || true; } \
+    | wc -l \
+    | tr -d '[:space:]'
+)
+
+if [ "${OPEN_PRS}" -eq 0 ]; then
+  log "SKIP: No open PRs matching branch patterns (${BRANCH_PATTERNS_RAW})"
+  emit_result "skip_no_open_prs"
+  exit 0
+fi
+
+REPO=$(gh repo view --json nameWithOwner --jq '.nameWithOwner' 2>/dev/null || echo "")
+
+# Collect PRs that need QA
+PRS_NEEDING_QA=""
+QA_NEEDED=0
+
+while IFS=$'\t' read -r pr_number pr_branch pr_title pr_labels; do
+  if [ -z "${pr_number}" ] || [ -z "${pr_branch}" ]; then
+    continue
+  fi
+
+  # Filter by branch pattern
+  if ! printf '%s\n' "${pr_branch}" | grep -Eq "${BRANCH_REGEX}"; then
+    continue
+  fi
+
+  # Skip PRs with the skip label
+  if echo "${pr_labels}" | grep -q "${SKIP_LABEL}"; then
+    log "SKIP-QA: PR #${pr_number} (${pr_branch}) has '${SKIP_LABEL}' label"
+    continue
+  fi
+
+  # Skip PRs with [skip-qa] in their title
+  if echo "${pr_title}" | grep -qi '\[skip-qa\]'; then
+    log "SKIP-QA: PR #${pr_number} (${pr_branch}) has [skip-qa] in title"
+    continue
+  fi
+
+  # Skip PRs that already have a QA comment (idempotency)
+  ALL_COMMENTS=$(
+    {
+      gh pr view "${pr_number}" --json comments --jq '.comments[].body' 2>/dev/null || true
+      if [ -n "${REPO}" ]; then
+        gh api "repos/${REPO}/issues/${pr_number}/comments" --jq '.[].body' 2>/dev/null || true
+      fi
+    } | sort -u
+  )
+  if echo "${ALL_COMMENTS}" | grep -q '<!-- night-watch-qa-marker -->'; then
+    log "SKIP-QA: PR #${pr_number} (${pr_branch}) already has QA comment"
+    continue
+  fi
+
+  QA_NEEDED=1
+  PRS_NEEDING_QA="${PRS_NEEDING_QA} #${pr_number}"
+done < <(
+  echo "${PR_JSON}" \
+    | jq -r '.[] | [.number, .headRefName, .title, ([.labels[].name] | join(","))] | @tsv' 2>/dev/null || true
+)
+
+if [ "${QA_NEEDED}" -eq 0 ]; then
+  log "SKIP: All ${OPEN_PRS} open PR(s) matching patterns already have QA comments"
+  emit_result "skip_all_qa_done"
+  exit 0
+fi
+
+PRS_NEEDING_QA=$(echo "${PRS_NEEDING_QA}" \
+  | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]][[:space:]]*/ /g' -e 's/[[:space:]]*$//')
+PRS_NEEDING_QA_CSV="${PRS_NEEDING_QA// /,}"
+
+if [ -n "${NW_DEFAULT_BRANCH:-}" ]; then
+  DEFAULT_BRANCH="${NW_DEFAULT_BRANCH}"
+else
+  DEFAULT_BRANCH=$(detect_default_branch "${PROJECT_DIR}")
+fi
+QA_WORKTREE_DIR="$(dirname "${PROJECT_DIR}")/${PROJECT_NAME}-nw-qa-runner"
+
+log "START: Found PR(s) needing QA:${PRS_NEEDING_QA}"
+
+cleanup_worktrees "${PROJECT_DIR}"
+
+# Dry-run mode: print diagnostics and exit
+if [ "${NW_DRY_RUN:-0}" = "1" ]; then
+  echo "=== Dry Run: QA Runner ==="
+  echo "Provider: ${PROVIDER_CMD}"
+  echo "Branch Patterns: ${BRANCH_PATTERNS_RAW}"
+  echo "Skip Label: ${SKIP_LABEL}"
+  echo "QA Artifacts: ${QA_ARTIFACTS}"
+  echo "Auto-install Playwright: ${QA_AUTO_INSTALL_PLAYWRIGHT}"
+  echo "Open PRs needing QA:${PRS_NEEDING_QA}"
+  echo "Default Branch: ${DEFAULT_BRANCH}"
+  echo "QA Worktree: ${QA_WORKTREE_DIR}"
+  echo "Timeout: ${MAX_RUNTIME}s"
+  exit 0
+fi
+
+EXIT_CODE=0
+
+# Process each PR that needs QA
+for pr_ref in ${PRS_NEEDING_QA}; do
+  pr_num="${pr_ref#\#}"
+
+  cleanup_worktrees "${PROJECT_DIR}"
+  if ! prepare_detached_worktree "${PROJECT_DIR}" "${QA_WORKTREE_DIR}" "${DEFAULT_BRANCH}" "${LOG_FILE}"; then
+    log "FAIL: Unable to create isolated QA worktree ${QA_WORKTREE_DIR} for PR #${pr_num}"
+    EXIT_CODE=1
+    break
+  fi
+
+  log "QA: Checking out PR #${pr_num} in worktree"
+  if ! (cd "${QA_WORKTREE_DIR}" && gh pr checkout "${pr_num}" >> "${LOG_FILE}" 2>&1); then
+    log "WARN: Failed to checkout PR #${pr_num}, skipping"
+    EXIT_CODE=1
+    cleanup_worktrees "${PROJECT_DIR}"
+    continue
+  fi
+
+  case "${PROVIDER_CMD}" in
+    claude)
+      if (
+        cd "${QA_WORKTREE_DIR}" && timeout "${MAX_RUNTIME}" \
+          claude -p "/night-watch-qa" \
+            --dangerously-skip-permissions \
+            >> "${LOG_FILE}" 2>&1
+      ); then
+        log "QA: PR #${pr_num} — provider completed successfully"
+      else
+        local_exit=$?
+        log "QA: PR #${pr_num} — provider exited with code ${local_exit}"
+        if [ ${local_exit} -eq 124 ]; then
+          EXIT_CODE=124
+          break
+        fi
+        EXIT_CODE=${local_exit}
+      fi
+      ;;
+    codex)
+      if (
+        cd "${QA_WORKTREE_DIR}" && timeout "${MAX_RUNTIME}" \
+          codex --quiet \
+            --yolo \
+            --prompt "$(cat "${QA_WORKTREE_DIR}/.claude/commands/night-watch-qa.md")" \
+            >> "${LOG_FILE}" 2>&1
+      ); then
+        log "QA: PR #${pr_num} — provider completed successfully"
+      else
+        local_exit=$?
+        log "QA: PR #${pr_num} — provider exited with code ${local_exit}"
+        if [ ${local_exit} -eq 124 ]; then
+          EXIT_CODE=124
+          break
+        fi
+        EXIT_CODE=${local_exit}
+      fi
+      ;;
+    *)
+      log "ERROR: Unknown provider: ${PROVIDER_CMD}"
+      exit 1
+      ;;
+  esac
+
+  cleanup_worktrees "${PROJECT_DIR}"
+done
+
+cleanup_worktrees "${PROJECT_DIR}"
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+  log "DONE: QA runner completed successfully"
+  emit_result "success_qa" "prs=${PRS_NEEDING_QA_CSV}"
+elif [ ${EXIT_CODE} -eq 124 ]; then
+  log "TIMEOUT: QA runner killed after ${MAX_RUNTIME}s"
+  emit_result "timeout" "prs=${PRS_NEEDING_QA_CSV}"
+else
+  log "FAIL: QA runner exited with code ${EXIT_CODE}"
+  emit_result "failure" "prs=${PRS_NEEDING_QA_CSV}"
+fi
+
+exit "${EXIT_CODE}"

package/templates/night-watch-qa.md

@@ -0,0 +1,157 @@
+You are the Night Watch QA agent. Your job is to analyze open PRs, generate appropriate tests for the changes, run them, and report results with visual evidence.
+
+## Context
+
+You are running inside a worktree checked out to a PR branch. Your goal is to:
+1. Analyze what changed in this PR compared to the base branch
+2. Determine if the changes are UI-related, API-related, or both
+3. Generate appropriate tests (Playwright e2e for UI, integration tests for API)
+4. Run the tests and capture artifacts (screenshots, videos for UI)
+5. Commit the tests and artifacts, then comment on the PR with results
+
+## Environment Variables Available
+- `NW_QA_ARTIFACTS` — What to capture: "screenshot", "video", or "both" (default: "both")
+- `NW_QA_AUTO_INSTALL_PLAYWRIGHT` — "1" to auto-install Playwright if missing
+
+## Instructions
+
+### Step 1: Analyze the PR diff
+
+Get the diff against the base branch:
+```
+git diff origin/${DEFAULT_BRANCH}...HEAD --name-only
+git diff origin/${DEFAULT_BRANCH}...HEAD --stat
+```
+
+Read the changed files to understand what the PR introduces.
+
+### Step 2: Classify and Decide
+
+Based on the diff, determine:
+- **UI changes**: New/modified components, pages, layouts, styles, client-side logic
+- **API changes**: New/modified endpoints, controllers, services, middleware, database queries
+- **Both**: PR touches both UI and API code
+- **No tests needed**: Trivial changes (docs, config, comments only) — in this case, post a comment saying "QA: No tests needed for this PR" and stop
+
+### Step 3: Prepare Test Infrastructure
+
+**For UI tests (Playwright):**
+1. Check if Playwright is available: `npx playwright --version`
+2. If not available and `NW_QA_AUTO_INSTALL_PLAYWRIGHT=1`:
+   - Run `npm install -D @playwright/test` (or yarn/pnpm equivalent based on lockfile)
+   - Run `npx playwright install chromium`
+3. If not available and auto-install is disabled, skip UI tests and note in the report
+
+**For API tests:**
+- Use the project's existing test framework (vitest, jest, or mocha — detect from package.json)
+- If no test framework exists, use vitest
+
+### Step 4: Generate Tests
+
+**UI Tests (Playwright):**
+- Create test files in `tests/e2e/qa/` (or the project's existing e2e directory)
+- Test the specific feature/page changed in the PR
+- Configure Playwright for artifacts based on `NW_QA_ARTIFACTS`:
+  - `"screenshot"`: `screenshot: 'on'` only
+  - `"video"`: `video: { mode: 'on', size: { width: 1280, height: 720 } }` only
+  - `"both"`: Both screenshot and video enabled
+- Name test files with a `qa-` prefix: `qa-<feature-name>.spec.ts`
+- Include at minimum: navigation to the feature, interaction with key elements, visual assertions
+
+**API Tests:**
+- Create test files in `tests/integration/qa/` (or the project's existing test directory)
+- Test the specific endpoints changed in the PR
+- Include: happy path, error cases, validation checks
+- Name test files with a `qa-` prefix: `qa-<endpoint-name>.test.ts`
+
+### Step 5: Run Tests
+
+**UI Tests:**
+```bash
+npx playwright test tests/e2e/qa/ --reporter=list
+```
+
+**API Tests:**
+```bash
+npx vitest run tests/integration/qa/ --reporter=verbose
+# (or equivalent for the project's test runner)
+```
+
+Capture the test output for the report.
+
+### Step 6: Collect Artifacts
+
+Move Playwright artifacts (screenshots, videos) to `qa-artifacts/` in the project root:
+```bash
+mkdir -p qa-artifacts
+# Copy from playwright-report/ or test-results/ to qa-artifacts/
+```
+
+### Step 7: Commit and Push
+
+```bash
+git add tests/e2e/qa/ tests/integration/qa/ qa-artifacts/ || true
+git add -A tests/*/qa/ qa-artifacts/ || true
+git commit -m "test(qa): add automated QA tests for PR changes
+
+- Generated by Night Watch QA agent
+- <UI tests: X passing, Y failing | No UI tests>
+- <API tests: X passing, Y failing | No API tests>
+- Artifacts: <screenshots, videos | screenshots | videos | none>
+
+Co-Authored-By: Claude <noreply@anthropic.com>"
+git push origin HEAD
+```
+
+### Step 8: Comment on PR
+
+Post a comment on the PR with results. Use the `<!-- night-watch-qa-marker -->` HTML comment for idempotency detection.
+
+```bash
+gh pr comment <PR_NUMBER> --body "<!-- night-watch-qa-marker -->
+## Night Watch QA Report
+
+### Changes Classification
+- **Type**: <UI | API | UI + API>
+- **Files changed**: <count>
+
+### Test Results
+
+<If UI tests>
+#### UI Tests (Playwright)
+- **Status**: <All passing | X of Y failing>
+- **Tests**: <count> test(s) in <count> file(s)
+
+<If screenshots captured>
+#### Screenshots
+<For each screenshot>
+![<description>](../blob/<branch>/qa-artifacts/<filename>)
+</For>
+</If>
+
+<If video captured>
+#### Video Recording
+Video artifact committed to \`qa-artifacts/\` — view in the PR's file changes.
+</If>
+</If>
+
+<If API tests>
+#### API Tests
+- **Status**: <All passing | X of Y failing>
+- **Tests**: <count> test(s) in <count> file(s)
+</If>
+
+<If no tests generated>
+**QA: No tests needed for this PR** — changes are trivial (docs, config, comments).
+</If>
+
+---
+*Night Watch QA Agent*"
+```
+
+### Important Rules
+- Process each PR **once** per run. Do NOT loop or retry after pushing.
+- Do NOT modify existing project tests — only add new files in `qa/` subdirectories.
+- If tests fail, still commit and report — the failures are useful information.
+- Keep test files self-contained and independent from each other.
+- Follow the project's existing code style and conventions (check CLAUDE.md, package.json scripts, tsconfig).

package/templates/night-watch.config.json

@@ -13,5 +13,18 @@
   "minReviewScore": 80,
   "maxLogSize": 524288,
   "cronSchedule": "0 0-21 * * *",
-  "reviewerSchedule": "0 0,3,6,9,12,15,18,21 * * *"
+  "reviewerSchedule": "0 0,3,6,9,12,15,18,21 * * *",
+  "autoMerge": false,
+  "autoMergeMethod": "squash",
+  "fallbackOnRateLimit": false,
+  "claudeModel": "sonnet",
+  "qa": {
+    "enabled": true,
+    "schedule": "30 1,7,13,19 * * *",
+    "maxRuntime": 3600,
+    "branchPatterns": [],
+    "artifacts": "both",
+    "skipLabel": "skip-qa",
+    "autoInstallPlaywright": true
+  }
 }