#!/usr/bin/env bash
set -euo pipefail

ROOT=/usr/local/valherun/pxe_boot_alma
WATCH_ROOT=/tmp/probe-stage-watch
SUP_ROOT="$WATCH_ROOT/supervisor"
LOCK_FILE="$SUP_ROOT/claude.lock"
RUN_LOG="$SUP_ROOT/claude-runs.log"
PROMPT_FILE="$SUP_ROOT/claude-prompt.txt"
LAST_MSG_FILE="$SUP_ROOT/claude-last-message.txt"
STDOUT_FILE="$SUP_ROOT/claude-stdout.txt"
LATEST_FILE="$WATCH_ROOT/latest-summary.txt"
STATE_FILE="$WATCH_ROOT/state.env"
FAIL_LOG="$WATCH_ROOT/failures.md"
MEMORY_FILE="$ROOT/codex-running-memory.md"
TRIAGE_FILE="$WATCH_ROOT/needs-triage.env"
CLAUDE_BIN="${CLAUDE_BIN:-/home/valherun/.local/bin/claude}"
export PATH="/home/valherun/.local/bin:$PATH"

mkdir -p "$SUP_ROOT"

timestamp() {
  date '+%F %T'
}

latest_attempt_dir() {
  find "$WATCH_ROOT" -maxdepth 1 -type d -name '20*-attempt-*' | sort | tail -n 1
}

print_triage_status() {
  if [[ -f "$TRIAGE_FILE" ]]; then
    cat "$TRIAGE_FILE"
    return
  fi

  if [[ -f "$STATE_FILE" ]]; then
    # shellcheck disable=SC1090
    source "$STATE_FILE"
  fi

  echo "no active triage request"
  echo "watcher_phase='${CURRENT_PHASE:-unknown}'"
  echo "watcher_detail='${CURRENT_DETAIL:-}'"
}

{
  flock -n 9 || exit 0

  attempt_dir="$(latest_attempt_dir || true)"
  {
    echo "You are a failure triage and repair assistant for a PXE/kickstart harness."
    echo "Diagnose likely root cause from the latest failure."
    echo "When the cause is clear and the fix is small and safe, edit files directly."
    echo "Prefer the smallest safe mitigation."
    echo "Update $MEMORY_FILE with a short bullet when you learn or fix something important."
    echo "Do not remove scripts permanently from the repo."
    echo
    echo "Time: $(timestamp)"
    echo
    echo "Watcher summary:"
    [[ -f "$LATEST_FILE" ]] && cat "$LATEST_FILE"
    echo
    echo "Watcher state:"
    [[ -f "$STATE_FILE" ]] && cat "$STATE_FILE"
    echo
    echo "Triage request:"
    print_triage_status
    echo
    echo "Failure ledger tail:"
    [[ -f "$FAIL_LOG" ]] && tail -n 40 "$FAIL_LOG"
    echo
    echo "Running memory tail:"
    [[ -f "$MEMORY_FILE" ]] && tail -n 80 "$MEMORY_FILE"
    echo
    echo "Latest attempt dir: ${attempt_dir:-missing}"
    if [[ -n "${attempt_dir:-}" ]]; then
      echo
      echo "Latest attempt log:"
      sed -n '1,240p' "$attempt_dir/attempt.log" 2>/dev/null || true
      echo
      echo "Latest attempt files:"
      find "$attempt_dir" -maxdepth 1 -type f | sort
      echo
      [[ -f "$attempt_dir/stage-meta.txt" ]] && { echo "stage-meta:"; cat "$attempt_dir/stage-meta.txt"; echo; }
      [[ -f "$attempt_dir/included-scripts.txt" ]] && { echo "included-scripts:"; cat "$attempt_dir/included-scripts.txt"; echo; }
      [[ -f "$attempt_dir/boot.log" ]] && { echo "boot.log tail:"; tail -n 200 "$attempt_dir/boot.log"; echo; }
      [[ -f "$attempt_dir/ks-post.log" ]] && { echo "ks-post.log tail:"; tail -n 200 "$attempt_dir/ks-post.log"; echo; }
    fi
    echo
    echo "Reply with concise findings ordered by confidence. Include: likely cause, most suspicious files/lines, what you changed if you edited anything, minimal safe mitigation, and whether to retry immediately or wait for a repo fix."
  } > "$PROMPT_FILE"

  echo "[$(timestamp)] claude-supervisor start" >> "$RUN_LOG"
  if "$CLAUDE_BIN" -p --dangerously-skip-permissions \
      --allowedTools 'Read,Edit,Bash(grep:*),Bash(sed:*),Bash(cat:*),Bash(head:*),Bash(tail:*),Bash(rg:*),Bash(ls:*),Bash(git status:*),Bash(git diff:*)' \
      < "$PROMPT_FILE" > "$STDOUT_FILE" 2>&1; then
    cp "$STDOUT_FILE" "$LAST_MSG_FILE"
    status="ok"
  else
    status="failed"
  fi

  {
    echo "[$(timestamp)] claude-supervisor status=$status"
    echo "--- last message ---"
    cat "$LAST_MSG_FILE" 2>/dev/null || true
    echo
  } >> "$RUN_LOG"

} 9>"$LOCK_FILE"