feat(launchd): durable LaunchAgent for codex_session daemon

launchd/README.md

@@ -0,0 +1,55 @@
+# launchd — codex tab status daemon
+
+Durable supervision for `scripts/codex_session.py --daemon` via macOS launchd.
+Mirrors the pattern in `~/code/home-lab/launchd/` (conductor-tick).
+
+## Files
+
+| File | Purpose |
+|---|---|
+| `com.local.codex-tab-status.plist` | LaunchAgent template — `__REPO_ROOT__`, `__HOME__`, `__PYTHON3__` substituted at install time |
+| `install-codex-tab-status-launchd.sh` | Resolves python3, expands the plist, bootstraps into `gui/$UID`, verifies first tick |
+| `uninstall-codex-tab-status-launchd.sh` | Idempotent bootout + plist removal |
+
+## Install
+
+```bash
+bash launchd/install-codex-tab-status-launchd.sh
+```
+
+Verifies by:
+- Tailing the log at `~/.cache/codex-tab-status.log`
+- Listing existing codex signal files at `~/.cache/claude-tab-status/codex-*.json`
+
+After install, open a codex tab in iTerm2 — its title should pick up `⚡` while
+codex is working and `💤` once a turn completes. The upstream iTerm2 adapter
+(`claude_tab_status.py`) renders these unchanged.
+
+## Uninstall
+
+```bash
+bash launchd/uninstall-codex-tab-status-launchd.sh
+```
+
+Stops the daemon and removes the agent plist. Signal files in
+`~/.cache/claude-tab-status/` self-clean via the adapter's PID-liveness check
+when the codex processes they describe exit.
+
+## Manage
+
+| Action | Command |
+|---|---|
+| Status | `launchctl print gui/$UID/com.local.codex-tab-status` |
+| Logs | `tail -f ~/.cache/codex-tab-status.log` |
+| Force restart | `launchctl kickstart -k gui/$UID/com.local.codex-tab-status` |
+
+## Why `KeepAlive` not `StartInterval`
+
+`codex_session.py --daemon` is a long-running poll loop (it watches for codex
+process start/exit + scans rollout JSONLs continuously). That's a "keep this
+running, respawn if it dies" job, not a "fire this every N seconds" job. So
+the plist uses `KeepAlive=true` without `StartInterval`.
+
+Compare conductor-tick (`com.mikebook.conductor-tick`) which IS one-shot
+per 120s — that uses `StartInterval=120` without `KeepAlive`. Different
+script lifecycles, different launchd primitives.

launchd/com.local.codex-tab-status.plist

@@ -0,0 +1,54 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+  <key>Label</key>
+  <string>com.local.codex-tab-status</string>
+
+  <!--
+    Codex tab status signal producer.
+
+    Long-running daemon that writes JSON signal files to
+    ~/.cache/claude-tab-status/codex-<rollout-uuid>.json describing each
+    live codex session's state. The upstream iTerm2 adapter
+    (claude_tab_status.py) consumes these files and renders the same
+    ⚡ / 💤 prefix it already shows for claude tabs.
+
+    KeepAlive=true: the daemon is a long-running poll loop, not an
+    interval-driven one-shot. If it crashes, launchd respawns it.
+    Pattern mirrors com.mikebook.conductor-tick (home-lab launchd/),
+    minus the StartInterval (this is daemonized, not tick-fired).
+
+    Install via launchd/install-codex-tab-status-launchd.sh.
+  -->
+
+  <key>ProgramArguments</key>
+  <array>
+    <string>__PYTHON3__</string>
+    <string>__REPO_ROOT__/scripts/codex_session.py</string>
+    <string>--daemon</string>
+  </array>
+
+  <key>RunAtLoad</key>
+  <true/>
+
+  <key>KeepAlive</key>
+  <true/>
+
+  <key>StandardOutPath</key>
+  <string>__HOME__/.cache/codex-tab-status.log</string>
+  <key>StandardErrorPath</key>
+  <string>__HOME__/.cache/codex-tab-status.log</string>
+
+  <key>WorkingDirectory</key>
+  <string>__HOME__</string>
+
+  <key>EnvironmentVariables</key>
+  <dict>
+    <key>HOME</key>
+    <string>__HOME__</string>
+    <key>PATH</key>
+    <string>/opt/homebrew/bin:/usr/bin:/bin</string>
+  </dict>
+</dict>
+</plist>

launchd/install-codex-tab-status-launchd.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# Install com.local.codex-tab-status as a launchd LaunchAgent.
+#
+# Template-expands the plist into ~/Library/LaunchAgents, bootstraps into
+# gui/$UID domain, enables, verifies. Uses modern launchctl bootstrap/
+# enable/print verbs (NOT the deprecated load/unload).
+#
+# Pattern source: ~/code/home-lab/launchd/install-conductor-tick-launchd.sh
+set -euo pipefail
+
+SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" &>/dev/null && pwd)"
+REPO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+LAUNCH_AGENTS_DIR="${HOME}/Library/LaunchAgents"
+PLIST_NAME="com.local.codex-tab-status.plist"
+LABEL="com.local.codex-tab-status"
+SRC="${SCRIPT_DIR}/${PLIST_NAME}"
+DST="${LAUNCH_AGENTS_DIR}/${PLIST_NAME}"
+LOG_PATH="${HOME}/.cache/codex-tab-status.log"
+
+# Resolve a real python3 — prefer homebrew, fall back to system. The plist
+# embeds the resolved path so launchd doesn't need to search PATH at fire.
+PYTHON3="$(command -v python3 || true)"
+if [[ -x /opt/homebrew/bin/python3 ]]; then
+  PYTHON3="/opt/homebrew/bin/python3"
+fi
+if [[ -z "${PYTHON3}" || ! -x "${PYTHON3}" ]]; then
+  echo "[install] FATAL: no python3 found on PATH" >&2
+  exit 1
+fi
+
+echo "[install] python3:     ${PYTHON3}"
+echo "[install] repo root:   ${REPO_ROOT}"
+echo "[install] plist src:   ${SRC}"
+echo "[install] plist dst:   ${DST}"
+echo "[install] log path:    ${LOG_PATH}"
+
+mkdir -p "${LAUNCH_AGENTS_DIR}" "${HOME}/.cache"
+
+# Template-expand placeholders
+sed \
+  -e "s|__REPO_ROOT__|${REPO_ROOT}|g" \
+  -e "s|__HOME__|${HOME}|g" \
+  -e "s|__PYTHON3__|${PYTHON3}|g" \
+  "${SRC}" > "${DST}"
+
+echo "[install] bootout (idempotent) then bootstrap into gui/$(id -u)"
+launchctl bootout "gui/$(id -u)/${LABEL}" 2>/dev/null || true
+launchctl bootstrap "gui/$(id -u)" "${DST}"
+launchctl enable "gui/$(id -u)/${LABEL}"
+
+echo "[install] launchctl print:"
+launchctl print "gui/$(id -u)/${LABEL}" 2>/dev/null | sed -n '1,18p'
+
+echo
+echo "[install] waiting 3s for daemon to start..."
+sleep 3
+echo "[install] log tail (${LOG_PATH}):"
+tail -10 "${LOG_PATH}" 2>/dev/null || echo "  (log not yet created — that's fine if no codex tabs are open)"
+
+echo
+echo "[install] signal files in ~/.cache/claude-tab-status/:"
+ls -1 "${HOME}/.cache/claude-tab-status/codex-"*.json 2>/dev/null | head -5 \
+  || echo "  (none yet — open a codex tab in iTerm2 and re-check)"
+
+echo
+echo "[install] done — uninstall with: bash ${SCRIPT_DIR}/uninstall-codex-tab-status-launchd.sh"

launchd/uninstall-codex-tab-status-launchd.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+# Uninstall com.local.codex-tab-status.
+#
+# Idempotent: bootout succeeds even if not loaded; rm -f tolerates a missing
+# plist. Leaves signal files in ~/.cache/claude-tab-status/ alone — they
+# self-clean via the adapter's PID-liveness check when their codex process
+# exits.
+set -euo pipefail
+
+LAUNCH_AGENTS_DIR="${HOME}/Library/LaunchAgents"
+PLIST_NAME="com.local.codex-tab-status.plist"
+LABEL="com.local.codex-tab-status"
+DST="${LAUNCH_AGENTS_DIR}/${PLIST_NAME}"
+
+echo "[uninstall] bootout gui/$(id -u)/${LABEL}"
+launchctl bootout "gui/$(id -u)/${LABEL}" 2>/dev/null || true
+
+echo "[uninstall] rm -f ${DST}"
+rm -f "${DST}"
+
+echo "[uninstall] done — codex tabs will lose their status indicators next time the iTerm2 adapter sweeps."