feat: codex-run.sh watchdog wrapper + stub tests

scripts/codex-run.sh

@@ -0,0 +1,74 @@
+#!/usr/bin/env bash
+# codex-crew: deterministic wrapper around `codex exec` with stall watchdog.
+#
+# Usage:
+#   codex-run.sh <job_dir> exec   [codex-flags...] [prompt]
+#   codex-run.sh <job_dir> resume <thread_id> [codex-flags...] [prompt]
+#   codex-run.sh <job_dir> review [--base X|--commit Y|--uncommitted] [codex-flags...] [prompt]
+#
+# If <job_dir>/prompt.md exists and no positional prompt given, it is fed via stdin.
+# Review runs against the git repo at the CURRENT WORKING DIRECTORY (no -C support).
+#
+# Files written to <job_dir>: events.jsonl last.md err.log thread_id status
+# status: ok | stalled | error: <reason>.  Exit code 0 only on ok.
+#
+# Env knobs:
+#   CODEX_STALL_TIMEOUT  seconds of event silence before kill (default 600)
+#   CODEX_WATCH_INTERVAL watchdog poll interval seconds (default 15)
+#   CODEX_BIN            codex binary (default codex; tests use a stub)
+set -u
+
+JOB_DIR=$1; MODE=$2; shift 2
+mkdir -p "$JOB_DIR"
+JOB_DIR=$(cd "$JOB_DIR" && pwd)
+EV=$JOB_DIR/events.jsonl LAST=$JOB_DIR/last.md ERR=$JOB_DIR/err.log
+STALL=${CODEX_STALL_TIMEOUT:-600}
+TICK=${CODEX_WATCH_INTERVAL:-15}
+BIN=${CODEX_BIN:-codex}
+
+case $MODE in
+  exec)   CMD=("$BIN" exec --json --skip-git-repo-check -o "$LAST") ;;
+  resume) TID=$1; shift; printf '%s\n' "$TID" > "$JOB_DIR/thread_id"
+          CMD=("$BIN" exec resume "$TID" --json --skip-git-repo-check -o "$LAST") ;;
+  review) CMD=("$BIN" exec review --json -o "$LAST") ;;
+  *) echo "usage: codex-run.sh <job_dir> exec|resume|review ..." >&2; exit 2 ;;
+esac
+
+rm -f "$JOB_DIR/status" "$LAST"; : > "$EV"; : > "$ERR"
+
+if [[ $# -eq 0 && -s $JOB_DIR/prompt.md ]]; then
+  "${CMD[@]}" - < "$JOB_DIR/prompt.md" > "$EV" 2> "$ERR" &
+else
+  "${CMD[@]}" "$@" < /dev/null > "$EV" 2> "$ERR" &
+fi
+PID=$!
+
+( while kill -0 "$PID" 2>/dev/null; do
+    sleep "$TICK"
+    now=$(date +%s); mt=$(stat -c %Y "$EV" 2>/dev/null || echo "$now")
+    if (( now - mt > STALL )); then
+      echo stalled > "$JOB_DIR/status"
+      kill "$PID" 2>/dev/null; sleep 3; kill -9 "$PID" 2>/dev/null
+      break
+    fi
+  done ) &
+WATCH=$!
+
+wait "$PID"; RC=$?
+kill "$WATCH" 2>/dev/null; wait "$WATCH" 2>/dev/null
+
+TID_FOUND=$(grep -m1 -o '"thread_id":"[^"]*"' "$EV" | cut -d'"' -f4)
+[[ -n $TID_FOUND ]] && printf '%s\n' "$TID_FOUND" > "$JOB_DIR/thread_id"
+
+if [[ -s $JOB_DIR/status ]]; then
+  : # stalled, written by watchdog
+elif [[ $RC -eq 0 && -s $LAST ]]; then
+  echo ok > "$JOB_DIR/status"
+else
+  echo "error: exit=$RC $(tail -c 300 "$ERR" | tr '\n' ' ')" > "$JOB_DIR/status"
+fi
+
+ST=$(head -n1 "$JOB_DIR/status")
+echo "status: $ST"
+echo "thread_id: $(cat "$JOB_DIR/thread_id" 2>/dev/null || echo none)"
+[[ $ST == ok ]]