@jonit-dev/night-watch-cli 1.7.49 → 1.7.51

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

@@ -27,6 +27,12 @@ AUTO_MERGE_METHOD="${NW_AUTO_MERGE_METHOD:-squash}"
 TARGET_PR="${NW_TARGET_PR:-}"
 PARALLEL_ENABLED="${NW_REVIEWER_PARALLEL:-1}"
 WORKER_MODE="${NW_REVIEWER_WORKER_MODE:-0}"
+PRD_DIR_REL="${NW_PRD_DIR:-docs/PRDs/night-watch}"
+if [[ "${PRD_DIR_REL}" = /* ]]; then
+  PRD_DIR="${PRD_DIR_REL}"
+else
+  PRD_DIR="${PROJECT_DIR}/${PRD_DIR_REL}"
+fi
 
 # Retry configuration
 REVIEWER_MAX_RETRIES="${NW_REVIEWER_MAX_RETRIES:-2}"
@@ -124,6 +130,175 @@ append_csv() {
   fi
 }
 
+truncate_for_prompt() {
+  local text="${1:-}"
+  local limit="${2:-7000}"
+  if [ "${#text}" -le "${limit}" ]; then
+    printf "%s" "${text}"
+  else
+    printf '%s\n\n[truncated to %s chars]' "${text:0:${limit}}" "${limit}"
+  fi
+}
+
+extract_linked_issue_numbers() {
+  local body="${1:-}"
+  printf '%s\n' "${body}" \
+    | grep -Eoi '(close[sd]?|fix(e[sd])?|resolve[sd]?)[[:space:]]*:?[[:space:]]*#[0-9]+' \
+    | grep -Eo '[0-9]+' \
+    | awk '!seen[$0]++' || true
+}
+
+find_prd_file_by_branch() {
+  local branch_name="${1:-}"
+  local branch_slug="${branch_name#*/}"
+  local branch_number=""
+  local candidate_dirs=()
+  local candidate=""
+  local base_name=""
+  local dir=""
+
+  if [ -z "${branch_slug}" ]; then
+    branch_slug="${branch_name}"
+  fi
+  [ -z "${branch_slug}" ] && return 1
+
+  if [ -d "${PRD_DIR}" ]; then
+    candidate_dirs+=("${PRD_DIR}")
+  fi
+  if [ -d "${PRD_DIR}/done" ]; then
+    candidate_dirs+=("${PRD_DIR}/done")
+  fi
+  [ "${#candidate_dirs[@]}" -eq 0 ] && return 1
+
+  for dir in "${candidate_dirs[@]}"; do
+    if [ -f "${dir}/${branch_slug}.md" ]; then
+      printf "%s" "${dir}/${branch_slug}.md"
+      return 0
+    fi
+  done
+
+  branch_number=$(printf '%s' "${branch_slug}" | grep -oE '^[0-9]+' || true)
+  for dir in "${candidate_dirs[@]}"; do
+    while IFS= read -r candidate; do
+      [ -z "${candidate}" ] && continue
+      base_name=$(basename "${candidate}" .md)
+      if [[ "${base_name}" == "${branch_slug}"* ]] || [[ "${branch_slug}" == "${base_name}"* ]]; then
+        printf "%s" "${candidate}"
+        return 0
+      fi
+      if [ -n "${branch_number}" ] && [[ "${base_name}" == "${branch_number}-"* ]]; then
+        printf "%s" "${candidate}"
+        return 0
+      fi
+    done < <(find "${dir}" -maxdepth 1 -type f -name '*.md' 2>/dev/null | sort)
+  done
+
+  return 1
+}
+
+build_prd_context_for_pr() {
+  local pr_number="${1:?PR number required}"
+  local pr_payload=""
+  local pr_title=""
+  local pr_branch=""
+  local pr_body=""
+  local pr_url=""
+  local issue_context=""
+  local issue_count=0
+  local issue_number=""
+  local issue_payload=""
+  local issue_title=""
+  local issue_body=""
+  local issue_excerpt=""
+  local prd_file=""
+  local prd_payload=""
+  local prd_excerpt=""
+  local prd_rel_path=""
+  local section=""
+
+  pr_payload=$(gh pr view "${pr_number}" --json title,headRefName,body,url 2>/dev/null || true)
+  pr_title=$(printf '%s' "${pr_payload}" | jq -r '.title // ""' 2>/dev/null || echo "")
+  pr_branch=$(printf '%s' "${pr_payload}" | jq -r '.headRefName // ""' 2>/dev/null || echo "")
+  pr_body=$(printf '%s' "${pr_payload}" | jq -r '.body // ""' 2>/dev/null || echo "")
+  pr_url=$(printf '%s' "${pr_payload}" | jq -r '.url // ""' 2>/dev/null || echo "")
+
+  if [ -n "${pr_body}" ]; then
+    while IFS= read -r issue_number; do
+      [ -z "${issue_number}" ] && continue
+      issue_count=$((issue_count + 1))
+      if [ "${issue_count}" -gt 2 ]; then
+        break
+      fi
+
+      issue_payload=$(gh issue view "${issue_number}" --json title,body,url 2>/dev/null || true)
+      issue_title=$(printf '%s' "${issue_payload}" | jq -r '.title // ""' 2>/dev/null || echo "")
+      issue_body=$(printf '%s' "${issue_payload}" | jq -r '.body // ""' 2>/dev/null || echo "")
+      [ -z "${issue_body}" ] && continue
+
+      issue_excerpt=$(truncate_for_prompt "${issue_body}" 4500)
+      issue_context="${issue_context}${issue_context:+$'\n\n'}Issue #${issue_number}: ${issue_title}
+${issue_excerpt}"
+    done < <(extract_linked_issue_numbers "${pr_body}")
+  fi
+
+  if [ -z "${issue_context}" ] && [ -n "${pr_branch}" ]; then
+    prd_file=$(find_prd_file_by_branch "${pr_branch}" || true)
+    if [ -n "${prd_file}" ] && [ -f "${prd_file}" ]; then
+      prd_payload=$(cat "${prd_file}" 2>/dev/null || true)
+      if [ -n "${prd_payload}" ]; then
+        prd_excerpt=$(truncate_for_prompt "${prd_payload}" 4500)
+        if [[ "${prd_file}" == "${PROJECT_DIR}/"* ]]; then
+          prd_rel_path="${prd_file#${PROJECT_DIR}/}"
+        else
+          prd_rel_path="${prd_file}"
+        fi
+      fi
+    fi
+  fi
+
+  section="### PR #${pr_number}"
+  if [ -n "${pr_title}" ]; then
+    section="${section} — ${pr_title}"
+  fi
+  section="${section}
+- branch: ${pr_branch:-unknown}"
+  if [ -n "${pr_url}" ]; then
+    section="${section}
+- url: ${pr_url}"
+  fi
+
+  if [ -n "${issue_context}" ]; then
+    section="${section}
+- context source: linked GitHub issue body
+${issue_context}"
+  elif [ -n "${prd_excerpt}" ]; then
+    section="${section}
+- context source: ${prd_rel_path}
+${prd_excerpt}"
+  else
+    section="${section}
+- context source: not found"
+  fi
+
+  printf "%s" "${section}"
+}
+
+build_prd_context_prompt() {
+  local pr_number=""
+  local entry=""
+  local combined=""
+
+  for pr_number in "$@"; do
+    [ -z "${pr_number}" ] && continue
+    entry=$(build_prd_context_for_pr "${pr_number}")
+    [ -z "${entry}" ] && continue
+    combined="${combined}${combined:+$'\n\n'}${entry}"
+  done
+
+  [ -z "${combined}" ] && return 0
+  printf '\n\n## PRD Context\nUse this product context while reviewing and fixing PRs.\n%s\n' "${combined}"
+}
+
 # Extract the latest review score from PR comments
 # Returns empty string if no score found
 get_pr_score() {
@@ -143,6 +318,135 @@ get_pr_score() {
     | grep -oP '\d+(?=/100)' || echo ""
 }
 
+# Count failed CI checks for a PR.
+# Uses JSON fields when available (more reliable across check name/status formats),
+# then falls back to text parsing for older/mocked gh outputs.
+get_pr_failed_ci_checks() {
+  local pr_number="${1:?PR number required}"
+  local failed_count=""
+
+  failed_count="$(
+    gh pr checks "${pr_number}" --json bucket,state,conclusion --jq '
+      [ .[]
+        | (.bucket // "" | ascii_downcase) as $bucket
+        | (.state // "" | ascii_downcase) as $state
+        | (.conclusion // "" | ascii_downcase) as $conclusion
+        | select(
+            $bucket == "fail" or
+            $bucket == "cancel" or
+            $state == "failure" or
+            $state == "error" or
+            $state == "cancelled" or
+            $conclusion == "failure" or
+            $conclusion == "error" or
+            $conclusion == "cancelled" or
+            $conclusion == "timed_out" or
+            $conclusion == "action_required" or
+            $conclusion == "startup_failure" or
+            $conclusion == "stale"
+          )
+      ] | length
+    ' 2>/dev/null || true
+  )"
+
+  if [[ "${failed_count}" =~ ^[0-9]+$ ]]; then
+    echo "${failed_count}"
+    return 0
+  fi
+
+  failed_count=$(
+    gh pr checks "${pr_number}" 2>/dev/null \
+      | grep -Eci 'fail|error|cancel|timed[_ -]?out|action_required|startup_failure|stale' || true
+  )
+
+  if [[ "${failed_count}" =~ ^[0-9]+$ ]]; then
+    echo "${failed_count}"
+  else
+    echo "0"
+  fi
+}
+
+# Return a semicolon-separated summary of failing CI checks for a PR.
+# Format: "<check name> [state=<state>, conclusion=<conclusion>]"
+get_pr_failed_ci_summary() {
+  local pr_number="${1:?PR number required}"
+  local failed_summary=""
+
+  failed_summary="$(
+    gh pr checks "${pr_number}" --json name,bucket,state,conclusion --jq '
+      [ .[]
+        | (.bucket // "" | ascii_downcase) as $bucket
+        | (.state // "" | ascii_downcase) as $state
+        | (.conclusion // "" | ascii_downcase) as $conclusion
+        | select(
+            $bucket == "fail" or
+            $bucket == "cancel" or
+            $state == "failure" or
+            $state == "error" or
+            $state == "cancelled" or
+            $conclusion == "failure" or
+            $conclusion == "error" or
+            $conclusion == "cancelled" or
+            $conclusion == "timed_out" or
+            $conclusion == "action_required" or
+            $conclusion == "startup_failure" or
+            $conclusion == "stale"
+          )
+        | "\(.name // "unknown") [state=\(.state // "unknown"), conclusion=\(.conclusion // "unknown")]"
+      ] | join("; ")
+    ' 2>/dev/null || true
+  )"
+
+  if [ -n "${failed_summary}" ]; then
+    echo "${failed_summary}"
+    return 0
+  fi
+
+  # Fallback for older/mocked outputs where JSON fields aren't available.
+  failed_summary=$(
+    gh pr checks "${pr_number}" 2>/dev/null \
+      | grep -Ei 'fail|error|cancel|timed[_ -]?out|action_required|startup_failure|stale' \
+      | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]][[:space:]]*/ /g' \
+      | paste -sd '; ' - || true
+  )
+
+  echo "${failed_summary}"
+}
+
+# Clean up reviewer-managed worktrees.
+# - Always removes the caller's runner worktree when runner_scope is provided.
+# - Only non-worker/controller processes perform broad cleanup to avoid
+#   parallel workers deleting each other's active worktrees.
+cleanup_reviewer_worktrees() {
+  local runner_scope="${1:-}"
+
+  if [ -n "${runner_scope}" ]; then
+    cleanup_worktrees "${PROJECT_DIR}" "${runner_scope}"
+  fi
+
+  if [ "${WORKER_MODE}" = "1" ]; then
+    return 0
+  fi
+
+  # Remove per-PR reviewer worktrees created by prompts from older runs.
+  cleanup_worktrees "${PROJECT_DIR}" "${PROJECT_NAME}-nw-review-"
+
+  # Remove legacy reviewer worktree naming used in some older prompt variants.
+  local escaped_project_name
+  escaped_project_name=$(printf '%s\n' "${PROJECT_NAME}" | sed 's/[][(){}.^$*+?|\\/]/\\&/g')
+  git -C "${PROJECT_DIR}" worktree list --porcelain 2>/dev/null \
+    | grep '^worktree ' \
+    | awk '{print $2}' \
+    | while read -r wt; do
+        local wt_basename
+        wt_basename=$(basename "${wt}")
+        if printf '%s\n' "${wt_basename}" | grep -Eq "^${escaped_project_name}-pr-?[0-9]+$"; then
+          log "CLEANUP: Removing legacy reviewer worktree ${wt}"
+          git -C "${PROJECT_DIR}" worktree remove --force "${wt}" 2>/dev/null || true
+        fi
+      done || true
+}
+
 # Validate provider
 if ! validate_provider "${PROVIDER_CMD}"; then
   echo "ERROR: Unknown provider: ${PROVIDER_CMD}" >&2
@@ -221,9 +525,14 @@ while IFS=$'\t' read -r pr_number pr_branch; do
     continue
   fi
 
-  FAILED_CHECKS=$(gh pr checks "${pr_number}" 2>/dev/null | grep -ci 'fail' || true)
+  FAILED_CHECKS=$(get_pr_failed_ci_checks "${pr_number}")
   if [ "${FAILED_CHECKS}" -gt 0 ]; then
-    log "INFO: PR #${pr_number} (${pr_branch}) has ${FAILED_CHECKS} failed CI check(s)"
+    FAILED_SUMMARY=$(get_pr_failed_ci_summary "${pr_number}")
+    if [ -n "${FAILED_SUMMARY}" ]; then
+      log "INFO: PR #${pr_number} (${pr_branch}) has ${FAILED_CHECKS} failed CI check(s): ${FAILED_SUMMARY}"
+    else
+      log "INFO: PR #${pr_number} (${pr_branch}) has ${FAILED_CHECKS} failed CI check(s)"
+    fi
     NEEDS_WORK=1
     PRS_NEEDING_WORK="${PRS_NEEDING_WORK} #${pr_number}"
     continue
@@ -325,6 +634,10 @@ fi
 
 log "START: Found PR(s) needing work:${PRS_NEEDING_WORK}"
 
+# Remove stale reviewer worktrees from previous interrupted runs.
+# Worker processes skip broad cleanup to avoid parallel interference.
+cleanup_reviewer_worktrees
+
 # Convert "#12 #34" into ["12", "34"] for worker fan-out.
 PR_NUMBER_ARRAY=()
 for pr_token in ${PRS_NEEDING_WORK}; do
@@ -355,7 +668,14 @@ if [ -z "${TARGET_PR}" ] && [ "${WORKER_MODE}" != "1" ] && [ "${PARALLEL_ENABLED
   declare -a WORKER_PRS=()
   declare -a WORKER_OUTPUTS=()
 
+  WORKER_IDX=0
+  WORKER_STAGGER_DELAY="${NW_REVIEWER_WORKER_STAGGER:-60}"
   for pr_number in "${PR_NUMBER_ARRAY[@]}"; do
+    if [ "${WORKER_IDX}" -gt 0 ]; then
+      log "PARALLEL: Staggering worker launch by ${WORKER_STAGGER_DELAY}s (worker $((WORKER_IDX + 1))/${#PR_NUMBER_ARRAY[@]})"
+      sleep "${WORKER_STAGGER_DELAY}"
+    fi
+
     worker_output=$(mktemp "/tmp/night-watch-pr-reviewer-${PROJECT_RUNTIME_KEY}-pr-${pr_number}.XXXXXX")
     WORKER_OUTPUTS+=("${worker_output}")
     WORKER_PRS+=("${pr_number}")
@@ -370,6 +690,7 @@ if [ -z "${TARGET_PR}" ] && [ "${WORKER_MODE}" != "1" ] && [ "${PARALLEL_ENABLED
     worker_pid=$!
     WORKER_PIDS+=("${worker_pid}")
     log "PARALLEL: Worker PID ${worker_pid} started for PR #${pr_number}"
+    WORKER_IDX=$((WORKER_IDX + 1))
   done
 
   EXIT_CODE=0
@@ -439,6 +760,10 @@ if [ -z "${TARGET_PR}" ] && [ "${WORKER_MODE}" != "1" ] && [ "${PARALLEL_ENABLED
     fi
   done
 
+  # Parent/controller process cleans up any per-PR reviewer worktrees that
+  # worker runs may have left behind.
+  cleanup_reviewer_worktrees
+
   emit_final_status "${EXIT_CODE}" "${PRS_NEEDING_WORK_CSV}" "${AUTO_MERGED_PRS}" "${AUTO_MERGE_FAILED_PRS}" "${MAX_WORKER_ATTEMPTS}" "${MAX_WORKER_FINAL_SCORE}"
   exit 0
 fi
@@ -449,7 +774,7 @@ if [ -n "${TARGET_PR}" ]; then
 fi
 REVIEW_WORKTREE_DIR="$(dirname "${PROJECT_DIR}")/${REVIEW_WORKTREE_BASENAME}"
 
-cleanup_worktrees "${PROJECT_DIR}" "${REVIEW_WORKTREE_BASENAME}"
+cleanup_reviewer_worktrees "${REVIEW_WORKTREE_BASENAME}"
 
 # Dry-run mode: print diagnostics and exit
 if [ "${NW_DRY_RUN:-0}" = "1" ]; then
@@ -479,12 +804,59 @@ if ! prepare_detached_worktree "${PROJECT_DIR}" "${REVIEW_WORKTREE_DIR}" "${DEFA
   exit 1
 fi
 
+REVIEWER_PROMPT_PATH=$(resolve_instruction_path_with_fallback "${REVIEW_WORKTREE_DIR}" "pr-reviewer.md" "night-watch-pr-reviewer.md" || true)
+if [ -z "${REVIEWER_PROMPT_PATH}" ]; then
+  log "FAIL: Missing reviewer prompt file. Checked pr-reviewer.md/night-watch-pr-reviewer.md in instructions/, .claude/commands/, and bundled templates/"
+  emit_result "failure" "reason=missing_reviewer_prompt"
+  exit 1
+fi
+REVIEWER_PROMPT_BUNDLED_NAME="pr-reviewer.md"
+if [[ "${REVIEWER_PROMPT_PATH}" == */night-watch-pr-reviewer.md ]]; then
+  REVIEWER_PROMPT_BUNDLED_NAME="night-watch-pr-reviewer.md"
+fi
+REVIEWER_PROMPT_PATH=$(prefer_bundled_prompt_if_legacy_command "${REVIEW_WORKTREE_DIR}" "${REVIEWER_PROMPT_PATH}" "${REVIEWER_PROMPT_BUNDLED_NAME}")
+REVIEWER_PROMPT_BASE=$(cat "${REVIEWER_PROMPT_PATH}")
+REVIEWER_PROMPT_REF=$(instruction_ref_for_prompt "${REVIEW_WORKTREE_DIR}" "${REVIEWER_PROMPT_PATH}")
+log "INFO: Using reviewer prompt from ${REVIEWER_PROMPT_REF}"
+
 EXIT_CODE=0
 ATTEMPTS_MADE=1
 FINAL_SCORE=""
 TARGET_SCOPE_PROMPT=""
 if [ -n "${TARGET_PR}" ]; then
   TARGET_SCOPE_PROMPT=$'\n\n## Target Scope\n- Only process PR #'"${TARGET_PR}"$'.\n- Ignore all other PRs.\n- If this PR no longer needs work, stop immediately.\n'
+
+  TARGET_MERGE_STATE=$(gh pr view "${TARGET_PR}" --json mergeStateStatus --jq '.mergeStateStatus' 2>/dev/null || echo "UNKNOWN")
+  TARGET_FAILED_CHECKS=$(get_pr_failed_ci_summary "${TARGET_PR}")
+  TARGET_SCORE=$(get_pr_score "${TARGET_PR}")
+
+  TARGET_SCOPE_PROMPT+=$'\n## Preflight Data (from CLI)\n- mergeStateStatus: '"${TARGET_MERGE_STATE}"$'\n'
+  if [ -n "${TARGET_FAILED_CHECKS}" ]; then
+    TARGET_SCOPE_PROMPT+=$'- failing checks: '"${TARGET_FAILED_CHECKS}"$'\n'
+  else
+    TARGET_SCOPE_PROMPT+=$'- failing checks: none detected\n'
+  fi
+  if [ -n "${TARGET_SCORE}" ]; then
+    TARGET_SCOPE_PROMPT+=$'- latest review score: '"${TARGET_SCORE}"$'/100\n'
+  else
+    TARGET_SCOPE_PROMPT+=$'- latest review score: not found\n'
+  fi
+fi
+
+PRD_CONTEXT_PROMPT=""
+if [ -n "${TARGET_PR}" ]; then
+  PRD_CONTEXT_PROMPT=$(build_prd_context_prompt "${TARGET_PR}")
+elif [ "${#PR_NUMBER_ARRAY[@]}" -gt 0 ]; then
+  PRD_CONTEXT_PROMPT=$(build_prd_context_prompt "${PR_NUMBER_ARRAY[@]}")
+fi
+if [ -n "${PRD_CONTEXT_PROMPT}" ]; then
+  if [ -n "${TARGET_PR}" ]; then
+    log "INFO: Added PRD context for PR #${TARGET_PR}"
+  else
+    log "INFO: Added PRD context for ${#PR_NUMBER_ARRAY[@]} PR(s)"
+  fi
+else
+  log "WARN: No PRD context found for current reviewer scope"
 fi
 
 # ── Retry Loop for Targeted PR Review ──────────────────────────────────────────
@@ -518,13 +890,14 @@ for ATTEMPT in $(seq 1 "${TOTAL_ATTEMPTS}"); do
   fi
 
   log "RETRY: Starting attempt ${ATTEMPT}/${TOTAL_ATTEMPTS} (timeout: ${ATTEMPT_TIMEOUT}s)"
+  LOG_LINE_BEFORE=$(wc -l < "${LOG_FILE}" 2>/dev/null || echo 0)
+  REVIEWER_PROMPT="${REVIEWER_PROMPT_BASE}${TARGET_SCOPE_PROMPT}${PRD_CONTEXT_PROMPT}"
 
   case "${PROVIDER_CMD}" in
     claude)
-      CLAUDE_PROMPT="/night-watch-pr-reviewer${TARGET_SCOPE_PROMPT}"
       if (
         cd "${REVIEW_WORKTREE_DIR}" && timeout "${ATTEMPT_TIMEOUT}" \
-          claude -p "${CLAUDE_PROMPT}" \
+          claude -p "${REVIEWER_PROMPT}" \
             --dangerously-skip-permissions \
             >> "${LOG_FILE}" 2>&1
       ); then
@@ -534,12 +907,11 @@ for ATTEMPT in $(seq 1 "${TOTAL_ATTEMPTS}"); do
       fi
       ;;
     codex)
-      CODEX_PROMPT="$(cat "${REVIEW_WORKTREE_DIR}/.claude/commands/night-watch-pr-reviewer.md")${TARGET_SCOPE_PROMPT}"
       if (
         cd "${REVIEW_WORKTREE_DIR}" && timeout "${ATTEMPT_TIMEOUT}" \
           codex --quiet \
             --yolo \
-            --prompt "${CODEX_PROMPT}" \
+            --prompt "${REVIEWER_PROMPT}" \
             >> "${LOG_FILE}" 2>&1
       ); then
         EXIT_CODE=0
@@ -553,8 +925,16 @@ for ATTEMPT in $(seq 1 "${TOTAL_ATTEMPTS}"); do
       ;;
   esac
 
-  # If provider failed (non-zero exit), don't retry
+  # If provider failed (non-zero exit), check for rate limit before giving up
   if [ "${EXIT_CODE}" -ne 0 ]; then
+    if [ "${EXIT_CODE}" -ne 124 ] && \
+       check_rate_limited "${LOG_FILE}" "${LOG_LINE_BEFORE}" && \
+       [ -n "${TARGET_PR}" ] && \
+       [ "${ATTEMPT}" -lt "${TOTAL_ATTEMPTS}" ]; then
+      log "RATE-LIMITED: 429 detected for PR #${TARGET_PR} (attempt ${ATTEMPT}/${TOTAL_ATTEMPTS}), retrying in 120s..."
+      sleep 120
+      continue
+    fi
     log "RETRY: Provider exited with code ${EXIT_CODE}, not retrying"
     break
   fi
@@ -585,7 +965,7 @@ for ATTEMPT in $(seq 1 "${TOTAL_ATTEMPTS}"); do
   fi
 done
 
-cleanup_worktrees "${PROJECT_DIR}" "${REVIEW_WORKTREE_BASENAME}"
+cleanup_reviewer_worktrees "${REVIEW_WORKTREE_BASENAME}"
 
 # ── Auto-merge eligible PRs ─────────────────────────────────────────────────────
 # After the reviewer completes, check for PRs that are merge-ready and queue them

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

@@ -53,6 +53,90 @@ emit_result() {
   fi
 }
 
+decode_base64_value() {
+  local value="${1:-}"
+  if [ -z "${value}" ]; then
+    return 0
+  fi
+  if printf '%s' "${value}" | base64 --decode >/dev/null 2>&1; then
+    printf '%s' "${value}" | base64 --decode
+  else
+    printf '%s' "${value}" | base64 -d 2>/dev/null || true
+  fi
+}
+
+get_pr_comment_bodies_base64() {
+  local pr_number="${1:?PR number required}"
+  gh pr view "${pr_number}" --json comments --jq '.comments[]?.body | @base64' 2>/dev/null || true
+  if [ -n "${REPO:-}" ]; then
+    gh api "repos/${REPO}/issues/${pr_number}/comments" --jq '.[].body | @base64' 2>/dev/null || true
+  fi
+}
+
+get_latest_qa_comment_body() {
+  local pr_number="${1:?PR number required}"
+  local latest=""
+  local encoded=""
+  local decoded=""
+
+  while IFS= read -r encoded; do
+    [ -z "${encoded}" ] && continue
+    decoded=$(decode_base64_value "${encoded}")
+    if printf '%s' "${decoded}" | grep -q '<!-- night-watch-qa-marker -->'; then
+      latest="${decoded}"
+    fi
+  done < <(get_pr_comment_bodies_base64 "${pr_number}")
+
+  printf "%s" "${latest}"
+}
+
+pr_has_qa_generated_files() {
+  local pr_number="${1:?PR number required}"
+  gh pr view "${pr_number}" --json files --jq '.files[]?.path' 2>/dev/null \
+    | grep -Eq '^(qa-artifacts/|tests/.*/qa/)'
+}
+
+provider_output_looks_invalid() {
+  local from_line="${1:-0}"
+  if [ ! -f "${LOG_FILE}" ]; then
+    return 1
+  fi
+
+  tail -n "+$((from_line + 1))" "${LOG_FILE}" 2>/dev/null \
+    | grep -Eqi 'Unknown skill:|session is in a broken state|working directory .* no longer exists|Please restart this session'
+}
+
+validate_qa_evidence() {
+  local pr_number="${1:?PR number required}"
+  local qa_comment=""
+
+  qa_comment=$(get_latest_qa_comment_body "${pr_number}")
+  if [ -z "${qa_comment}" ]; then
+    log "FAIL-QA-EVIDENCE: PR #${pr_number} has no QA marker comment (<!-- night-watch-qa-marker -->)"
+    return 1
+  fi
+
+  if printf '%s' "${qa_comment}" | grep -Eqi 'QA: No tests needed for this PR|No tests needed'; then
+    return 0
+  fi
+
+  if ! pr_has_qa_generated_files "${pr_number}"; then
+    log "FAIL-QA-EVIDENCE: PR #${pr_number} has QA marker comment but no qa-artifacts/ or tests/*/qa/ files"
+    return 1
+  fi
+
+  if [ "${QA_ARTIFACTS}" = "screenshot" ] || [ "${QA_ARTIFACTS}" = "both" ]; then
+    if printf '%s' "${qa_comment}" | grep -q '#### UI Tests (Playwright)'; then
+      if ! printf '%s' "${qa_comment}" | grep -Eq '!\[[^]]*\]\([^)]*qa-artifacts/[^)]*\)'; then
+        log "FAIL-QA-EVIDENCE: PR #${pr_number} reports UI tests but comment lacks screenshot links to qa-artifacts/"
+        return 1
+      fi
+    fi
+  fi
+
+  return 0
+}
+
 # Validate provider
 if ! validate_provider "${PROVIDER_CMD}"; then
   echo "ERROR: Unknown provider: ${PROVIDER_CMD}" >&2
@@ -217,15 +301,32 @@ Artifacts: ${QA_ARTIFACTS}"
     continue
   fi
 
+  QA_PROMPT_PATH=$(resolve_instruction_path_with_fallback "${QA_WORKTREE_DIR}" "qa.md" "night-watch-qa.md" || true)
+  if [ -z "${QA_PROMPT_PATH}" ]; then
+    log "FAIL: Missing QA prompt file for PR #${pr_num}. Checked qa.md/night-watch-qa.md in instructions/, .claude/commands/, and bundled templates/"
+    EXIT_CODE=1
+    break
+  fi
+  QA_PROMPT_BUNDLED_NAME="qa.md"
+  if [[ "${QA_PROMPT_PATH}" == */night-watch-qa.md ]]; then
+    QA_PROMPT_BUNDLED_NAME="night-watch-qa.md"
+  fi
+  QA_PROMPT_PATH=$(prefer_bundled_prompt_if_legacy_command "${QA_WORKTREE_DIR}" "${QA_PROMPT_PATH}" "${QA_PROMPT_BUNDLED_NAME}")
+  QA_PROMPT=$(cat "${QA_PROMPT_PATH}")
+  QA_PROMPT_REF=$(instruction_ref_for_prompt "${QA_WORKTREE_DIR}" "${QA_PROMPT_PATH}")
+  log "QA: PR #${pr_num} — using prompt from ${QA_PROMPT_REF}"
+
+  LOG_LINE_BEFORE=$(wc -l < "${LOG_FILE}" 2>/dev/null || echo 0)
+  PROVIDER_OK=0
   case "${PROVIDER_CMD}" in
     claude)
       if (
         cd "${QA_WORKTREE_DIR}" && timeout "${MAX_RUNTIME}" \
-          claude -p "/night-watch-qa" \
+          claude -p "${QA_PROMPT}" \
             --dangerously-skip-permissions \
             >> "${LOG_FILE}" 2>&1
       ); then
-        log "QA: PR #${pr_num} — provider completed successfully"
+        PROVIDER_OK=1
       else
         local_exit=$?
         log "QA: PR #${pr_num} — provider exited with code ${local_exit}"
@@ -241,10 +342,10 @@ Artifacts: ${QA_ARTIFACTS}"
         cd "${QA_WORKTREE_DIR}" && timeout "${MAX_RUNTIME}" \
           codex --quiet \
             --yolo \
-            --prompt "$(cat "${QA_WORKTREE_DIR}/.claude/commands/night-watch-qa.md")" \
+            --prompt "${QA_PROMPT}" \
             >> "${LOG_FILE}" 2>&1
       ); then
-        log "QA: PR #${pr_num} — provider completed successfully"
+        PROVIDER_OK=1
       else
         local_exit=$?
         log "QA: PR #${pr_num} — provider exited with code ${local_exit}"
@@ -261,6 +362,17 @@ Artifacts: ${QA_ARTIFACTS}"
       ;;
   esac
 
+  if [ "${PROVIDER_OK}" -eq 1 ]; then
+    if provider_output_looks_invalid "${LOG_LINE_BEFORE}"; then
+      log "FAIL-QA-EVIDENCE: PR #${pr_num} provider output indicates an invalid automation run"
+      EXIT_CODE=1
+    elif ! validate_qa_evidence "${pr_num}"; then
+      EXIT_CODE=1
+    else
+      log "QA: PR #${pr_num} — provider completed with verifiable QA evidence"
+    fi
+  fi
+
   cleanup_worktrees "${PROJECT_DIR}"
 done