@aipper/aiws-spec 0.0.24 → 0.0.26

package/templates/workspace/tools/iflow_watchdog.sh

@@ -1,138 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-# iFlow unattended watchdog for AI Workspace (Arch Linux/systemd friendly).
-#
-# Goal:
-# - Run /server-drain in a loop (low output, small batches).
-# - If drain reports BLOCKED (exit 3) or crashes, run /server-fix in a fresh iFlow process (bounded).
-# - Keep outputs in .agentdocs/tmp/server-test/ to avoid blowing up interactive sessions.
-#
-# Safety:
-# - Intended for test environments only.
-# - Requires explicit IFLOW_WATCHDOG_YOLO=1 to run with --yolo.
-
-die() { echo "error: $*" >&2; exit 2; }
-need_file() { [[ -f "$1" ]] || die "missing: $1"; }
-have() { command -v "$1" >/dev/null 2>&1; }
-
-root="${IFLOW_WATCHDOG_ROOT:-.}"
-cd "$root"
-
-need_file "AI_WORKSPACE.md"
-need_file "REQUIREMENTS.md"
-need_file "tools/server_test_runner.py"
-need_file "secrets/test-accounts.json"
-
-have iflow || die "missing: iflow"
-have uv || die "missing: uv (runner requires uv)"
-
-csv="issues/server-api-issues.csv"
-workspace_dir="$(pwd)"
-log_dir="${IFLOW_WATCHDOG_LOG_DIR:-$workspace_dir/.agentdocs/tmp/server-test}"
-mkdir -p "$log_dir"
-
-fatal_log="$log_dir/watchdog-fatal.log"
-timestamp() { date +"%Y-%m-%dT%H:%M:%S%z"; }
-say() { printf "[%s] %s\n" "$(timestamp)" "$*"; }
-fatal() { say "fatal: $*" | tee -a "$fatal_log" >&2; exit 2; }
-
-yolo="${IFLOW_WATCHDOG_YOLO:-0}"
-if [[ "$yolo" != "1" ]]; then
-  fatal "unattended mode requires IFLOW_WATCHDOG_YOLO=1 (this will use iflow --yolo; test env only)"
-fi
-
-if ! iflow --version >/dev/null 2>&1; then
-  fatal "iflow command failed (check auth/config/environment)"
-fi
-
-date_suffix="$(date +%Y%m%d)"
-drain_log="${IFLOW_WATCHDOG_DRAIN_LOG:-$log_dir/iflow-drain-$date_suffix.log}"
-fix_log="${IFLOW_WATCHDOG_FIX_LOG:-$log_dir/iflow-fix-$date_suffix.log}"
-find "$log_dir" -maxdepth 1 -type f -name "iflow-*.log" -mtime +7 -delete 2>/dev/null || true
-
-lock_dir="$log_dir/.watchdog.lock"
-pid_file="$lock_dir/pid"
-if ! mkdir "$lock_dir" 2>/dev/null; then
-  other_pid="$(cat "$pid_file" 2>/dev/null || true)"
-  fatal "another watchdog is running (lock: $lock_dir pid=${other_pid:-unknown})"
-fi
-printf "%s\n" "$$" >"$pid_file" 2>/dev/null || true
-
-cleanup() {
-  say "cleanup: stopping child iflow processes (if any)" >>"$fatal_log" 2>/dev/null || true
-  pkill -P $$ iflow 2>/dev/null || true
-  rm -f "$pid_file" 2>/dev/null || true
-  rmdir "$lock_dir" 2>/dev/null || true
-}
-trap cleanup EXIT INT TERM
-
-drain_timeout="${IFLOW_WATCHDOG_DRAIN_TIMEOUT:-7200}"
-fix_timeout="${IFLOW_WATCHDOG_FIX_TIMEOUT:-3600}"
-drain_max_turns="${IFLOW_WATCHDOG_DRAIN_MAX_TURNS:-6}"
-fix_max_turns="${IFLOW_WATCHDOG_FIX_MAX_TURNS:-12}"
-drain_max_tokens="${IFLOW_WATCHDOG_DRAIN_MAX_TOKENS:-15000}"
-fix_max_tokens="${IFLOW_WATCHDOG_FIX_MAX_TOKENS:-30000}"
-
-max_cycles="${IFLOW_WATCHDOG_MAX_CYCLES:-200}"
-sleep_s="${IFLOW_WATCHDOG_SLEEP_S:-10}"
-fix_retry_max="${IFLOW_WATCHDOG_FIX_RETRY_MAX:-3}"
-fix_retry_file="$log_dir/.fix-retry-count"
-
-drain_max_endpoints="${IFLOW_DRAIN_MAX_ENDPOINTS:-30}"
-drain_sleep_s="${IFLOW_DRAIN_SLEEP_S:-0}"
-
-csv_has_blocked() { [[ -f "$csv" ]] && grep -qE ",BLOCKED," "$csv"; }
-
-run_drain() {
-  say "run: /server-drain (max_endpoints=$drain_max_endpoints)"
-  IFLOW_DRAIN_MAX_ENDPOINTS="$drain_max_endpoints" \
-  IFLOW_DRAIN_SLEEP_S="$drain_sleep_s" \
-  iflow --yolo -p "/server-drain" --timeout "$drain_timeout" --max-turns "$drain_max_turns" --max-tokens "$drain_max_tokens" >>"$drain_log" 2>&1
-}
-
-run_fix() {
-  say "run: /server-fix"
-  retry_count="$(cat "$fix_retry_file" 2>/dev/null || echo 0)"
-  if [[ "$retry_count" -ge "$fix_retry_max" ]]; then
-    fatal "fix reached retry limit ($fix_retry_max); check $fix_log"
-  fi
-  if iflow --yolo -p "/server-fix" --timeout "$fix_timeout" --max-turns "$fix_max_turns" --max-tokens "$fix_max_tokens" >>"$fix_log" 2>&1; then
-    rm -f "$fix_retry_file" 2>/dev/null || true
-  else
-    echo $((retry_count + 1)) >"$fix_retry_file" 2>/dev/null || true
-  fi
-}
-
-say "watchdog: start (cycles=$max_cycles sleep_s=$sleep_s)"
-say "logs: drain=$drain_log fix=$fix_log"
-
-cycle=1
-while [[ "$cycle" -le "$max_cycles" ]]; do
-  if csv_has_blocked; then
-    say "state: BLOCKED detected in $csv (trigger fix)"
-    run_fix
-  fi
-
-set +e
-run_drain
-rc=$?
-set -e
-
-if [[ "$rc" == "0" ]]; then
-  say "done: drain finished with all DONE/SKIP"
-  exit 0
-fi
-
-if [[ "$rc" == "3" ]] || csv_has_blocked; then
-  say "blocked: drain reported BLOCKED (rc=$rc), trigger fix"
-  run_fix
-else
-  say "warn: drain exited rc=$rc without BLOCKED; will retry"
-fi
-
-sleep "$sleep_s" || true
-cycle=$((cycle + 1))
-done
-
-fatal "reached max cycles ($max_cycles); check logs under $log_dir"

package/templates/workspace/tools/install_iflow_watchdog_systemd_user.sh

@@ -1,118 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-die() { echo "error: $*" >&2; exit 2; }
-have() { command -v "$1" >/dev/null 2>&1; }
-
-usage() {
-  cat <<'EOF'
-Usage:
-  bash tools/install_iflow_watchdog_systemd_user.sh
-  bash tools/install_iflow_watchdog_systemd_user.sh --name <instance>
-  bash tools/install_iflow_watchdog_systemd_user.sh --workspace <abs_path>
-  bash tools/install_iflow_watchdog_systemd_user.sh --name <instance> --workspace <abs_path>
-
-What it does:
-  - Installs systemd user instance units:
-      ~/.config/systemd/user/iflow-watchdog@.service
-      ~/.config/systemd/user/iflow-watchdog@.timer
-  - Writes per-instance env file:
-      ~/.config/iflow-watchdog/<instance>.env
-  - Enables and starts the timer:
-      systemctl --user enable --now iflow-watchdog@<instance>.timer
-
-Notes:
-  - You can run multiple instances on one server (different --name and --workspace).
-  - Unattended mode uses --yolo; only use in test environment.
-Defaults:
-  - --workspace defaults to current directory (pwd)
-  - --name defaults to basename(--workspace)
-EOF
-}
-
-name=""
-workspace=""
-while [[ $# -gt 0 ]]; do
-  case "$1" in
-    --name) name="${2:-}"; shift 2 ;;
-    --workspace) workspace="${2:-}"; shift 2 ;;
-    -h|--help) usage; exit 0 ;;
-    *) die "unknown arg: $1" ;;
-  esac
-done
-
-if [[ -z "$workspace" ]]; then
-  workspace="$(pwd)"
-fi
-
-if [[ -z "$name" ]]; then
-  name="$(basename "$workspace")"
-fi
-
-[[ "$workspace" == /* ]] || die "--workspace must be an absolute path"
-[[ -d "$workspace" ]] || die "workspace not found: $workspace"
-
-have systemctl || die "missing: systemctl"
-
-if [[ ! -f "$workspace/AI_WORKSPACE.md" ]] || [[ ! -f "$workspace/REQUIREMENTS.md" ]]; then
-  die "workspace must contain AI_WORKSPACE.md and REQUIREMENTS.md: $workspace"
-fi
-
-# Best-effort check that systemd --user is available (common pitfall on servers).
-if [[ -z "${XDG_RUNTIME_DIR:-}" ]] || ! systemctl --user is-system-running >/dev/null 2>&1; then
-  die "systemd --user not running (try: loginctl enable-linger $USER, then re-login)"
-fi
-
-unit_dir="$HOME/.config/systemd/user"
-env_dir="$HOME/.config/iflow-watchdog"
-mkdir -p "$unit_dir" "$env_dir"
-
-src_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-tpl_service="$src_dir/systemd/iflow-watchdog@.service"
-tpl_timer="$src_dir/systemd/iflow-watchdog@.timer"
-[[ -f "$tpl_service" ]] || die "missing template: $tpl_service"
-[[ -f "$tpl_timer" ]] || die "missing template: $tpl_timer"
-
-cp -a "$tpl_service" "$unit_dir/iflow-watchdog@.service"
-cp -a "$tpl_timer" "$unit_dir/iflow-watchdog@.timer"
-
-env_file="$env_dir/${name}.env"
-
-# Only allow unattended --yolo if AI_WORKSPACE.md declares test environment.
-yo="0"
-if grep -Eq '^\s*-\s*environment\s*:\s*"?test"?\s*$' "$workspace/AI_WORKSPACE.md"; then
-  yo="1"
-fi
-
-cat >"$env_file" <<EOF
-# Generated by tools/install_iflow_watchdog_systemd_user.sh
-WORKSPACE_DIR=${workspace}
-IFLOW_WATCHDOG_ROOT=.
-# Unattended mode (uses iflow --yolo). Only safe when environment=test.
-IFLOW_WATCHDOG_YOLO=${yo}
-# Optional: override logs dir (default: <workspace>/.agentdocs/tmp/server-test)
-# IFLOW_WATCHDOG_LOG_DIR=${workspace}/.agentdocs/tmp/server-test
-IFLOW_DRAIN_MAX_ENDPOINTS=30
-IFLOW_WATCHDOG_SLEEP_S=10
-EOF
-
-systemctl --user daemon-reload
-
-if [[ "$yo" == "1" ]]; then
-  systemctl --user enable --now "iflow-watchdog@${name}.timer"
-  enabled_msg="enabled"
-else
-  enabled_msg="not enabled (edit IFLOW_WATCHDOG_YOLO=1 after confirming test env)"
-fi
-
-echo "OK: installed iflow-watchdog@${name}"
-echo "env:  $env_file"
-echo "unit: $unit_dir/iflow-watchdog@.service"
-echo "timer: iflow-watchdog@${name}.timer ($enabled_msg)"
-echo "tune: edit $env_file, then run:"
-echo "  systemctl --user daemon-reload"
-echo "  systemctl --user restart iflow-watchdog@${name}.service"
-if [[ "$yo" != "1" ]]; then
-  echo "enable:"
-  echo "  systemctl --user enable --now iflow-watchdog@${name}.timer"
-fi

package/templates/workspace/tools/systemd/iflow-watchdog@.service

@@ -1,16 +0,0 @@
-[Unit]
-Description=iFlow AI Workspace API Test Watchdog (%i)
-After=network.target
-
-[Service]
-Type=simple
-# Per-instance environment file created by tools/install_iflow_watchdog_systemd_user.sh
-EnvironmentFile=%h/.config/iflow-watchdog/%i.env
-ExecStart=/usr/bin/env bash -lc 'cd "${WORKSPACE_DIR}" && bash tools/iflow_watchdog.sh'
-Restart=always
-RestartSec=30
-KillMode=mixed
-TimeoutStopSec=30
-
-[Install]
-WantedBy=default.target

package/templates/workspace/tools/systemd/iflow-watchdog@.timer

@@ -1,11 +0,0 @@
-[Unit]
-Description=Nightly iFlow API Test Watchdog (%i)
-
-[Timer]
-OnCalendar=*-*-* 02:00:00
-Persistent=true
-Unit=iflow-watchdog@%i.service
-
-[Install]
-WantedBy=timers.target
-