Harden mac-bridge Python pinning and add reboot auto-recover module.

Pin mac-bridge preset execution to a configurable venv interpreter (with mlx_lm preflight checks) and add a launchd-based runner self-heal flow so Mac reboots automatically restore bridge availability.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: 2 people

.github/workflows/mac-bridge.yaml

@@ -95,6 +95,10 @@ jobs:
       - name: Run preset (allowlist-validated executor)
         env:
           PYTHONPATH: .:sdks/python
+          # Pin the interpreter used by the bridge executor so runner
+          # reboots/PATH drift cannot silently fall back to a python
+          # without mlx/mlx_lm.
+          KAKEYA_MAC_PYTHON: ${{ vars.KAKEYA_MAC_PYTHON || '/Users/fluffy314/Documents/Kakeya-LLM-Inference-engine-pr109/.venv-mac/bin/python3' }}
           # Machine-local model locations come from the runner env,
           # never from the manifest (docs/ops/mac-m4-runner-setup.md).
           # Precedence: repo Actions variable > ~/kakeya-models/<name>
@@ -114,7 +118,9 @@ jobs:
           export KAKEYA_MAC_DRAFTER_ID="${KAKEYA_MAC_DRAFTER_ID_VAR:-z-lab/gemma-4-26B-A4B-it-DFlash}"
           export KAKEYA_MAC_FTHETA_DIR="${KAKEYA_MAC_FTHETA_DIR_VAR:-results/research/f_theta_v5_s5_sliding}"
           echo "verifier=$KAKEYA_MAC_VERIFIER_PATH"
-          python3 scripts/mac_bridge/run_preset.py \
+          "${KAKEYA_MAC_PYTHON}" --version
+          "${KAKEYA_MAC_PYTHON}" -c "import mlx_lm"
+          "${KAKEYA_MAC_PYTHON}" scripts/mac_bridge/run_preset.py \
             --manifest .mac-bridge/request.json
 
       - name: Commit results back to the request branch

scripts/mac_bridge/install_autorecover_launchagent.sh

@@ -0,0 +1,53 @@
+#!/usr/bin/env bash
+# Install a user LaunchAgent that re-checks the mac-bridge runner
+# after reboot and periodically self-heals it.
+
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+RECOVER_SCRIPT="${REPO_ROOT}/scripts/mac_bridge/recover_runner_after_reboot.sh"
+PLIST_DIR="${HOME}/Library/LaunchAgents"
+PLIST_PATH="${PLIST_DIR}/com.kakeya.mac-bridge-runner-autorecover.plist"
+LABEL="com.kakeya.mac-bridge-runner-autorecover"
+UID_NUM="$(id -u)"
+
+mkdir -p "$PLIST_DIR"
+[ -x "$RECOVER_SCRIPT" ] || chmod +x "$RECOVER_SCRIPT"
+
+cat >"$PLIST_PATH" <<EOF
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+  <key>Label</key>
+  <string>${LABEL}</string>
+
+  <key>ProgramArguments</key>
+  <array>
+    <string>/bin/bash</string>
+    <string>${RECOVER_SCRIPT}</string>
+  </array>
+
+  <key>WorkingDirectory</key>
+  <string>${REPO_ROOT}</string>
+
+  <key>RunAtLoad</key>
+  <true/>
+  <key>StartInterval</key>
+  <integer>60</integer>
+
+  <key>StandardOutPath</key>
+  <string>${HOME}/actions-runner/_diag/launchagent-autorecover.out.log</string>
+  <key>StandardErrorPath</key>
+  <string>${HOME}/actions-runner/_diag/launchagent-autorecover.err.log</string>
+</dict>
+</plist>
+EOF
+
+launchctl bootout "gui/${UID_NUM}" "${PLIST_PATH}" >/dev/null 2>&1 || true
+launchctl bootstrap "gui/${UID_NUM}" "${PLIST_PATH}"
+launchctl enable "gui/${UID_NUM}/${LABEL}" || true
+launchctl kickstart -k "gui/${UID_NUM}/${LABEL}" || true
+
+echo "[mac-bridge-autorecover] installed: ${PLIST_PATH}"
+echo "[mac-bridge-autorecover] label: ${LABEL}"

scripts/mac_bridge/recover_runner_after_reboot.sh

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+# Self-heal the GitHub Actions runner after Mac reboot.
+#
+# Idempotent:
+#  - if Runner.Listener is already alive: no-op
+#  - else try svc.sh start (when service is installed)
+#  - if service is not installed/usable, fall back to foreground run.sh
+#    via nohup so bridge jobs can still execute.
+
+set -euo pipefail
+
+RUNNER_DIR="${RUNNER_DIR:-$HOME/actions-runner}"
+LOG_DIR="${RUNNER_LOG_DIR:-$HOME/actions-runner/_diag}"
+mkdir -p "$LOG_DIR"
+
+log() { echo "[mac-bridge-recover] $*" >&2; }
+
+if pgrep -f "Runner.Listener.*${RUNNER_DIR}" >/dev/null 2>&1; then
+  log "Runner.Listener already running."
+  exit 0
+fi
+
+if [ ! -x "${RUNNER_DIR}/run.sh" ]; then
+  log "runner not found at ${RUNNER_DIR}; skipping."
+  exit 0
+fi
+
+if [ -x "${RUNNER_DIR}/svc.sh" ] && "${RUNNER_DIR}/svc.sh" status 2>/dev/null | grep -q "installed"; then
+  log "service installed; attempting svc.sh start"
+  if "${RUNNER_DIR}/svc.sh" start >/dev/null 2>&1; then
+    sleep 2
+    if pgrep -f "Runner.Listener.*${RUNNER_DIR}" >/dev/null 2>&1; then
+      log "runner started via service."
+      exit 0
+    fi
+  fi
+  log "service start did not bring up listener, falling back to run.sh"
+fi
+
+ts="$(date +%Y%m%d_%H%M%S)"
+nohup "${RUNNER_DIR}/run.sh" >"${LOG_DIR}/runner-nohup-${ts}.log" 2>&1 &
+sleep 2
+if pgrep -f "Runner.Listener.*${RUNNER_DIR}" >/dev/null 2>&1; then
+  log "runner started via nohup run.sh fallback."
+  exit 0
+fi
+
+log "failed to start runner listener."
+exit 1

scripts/mac_bridge/setup_mac.sh

@@ -173,6 +173,13 @@ else
     ok "skipped (rerun with --with-tailscale to enable M2 interactive SSH)"
 fi
 
+step "Runner auto-recover module (reboot self-heal)"
+chmod +x scripts/mac_bridge/recover_runner_after_reboot.sh \
+          scripts/mac_bridge/install_autorecover_launchagent.sh
+bash scripts/mac_bridge/install_autorecover_launchagent.sh \
+    || die "failed to install launchagent auto-recover module"
+ok "launchagent auto-recover installed (checks runner every 60s + RunAtLoad)"
+
 printf '\n\033[1m\033[32mMac bridge ready.\033[0m Any push to mac-bridge/** (or AgentMemory/mac-bridge-*) now executes here.\n'
 printf 'Smoke it from any clone with push rights:\n'
 printf '  PYTHONPATH=.:sdks/python python3 scripts/mac_bridge/kakeya_mac.py run --preset mlx-env-probe --wait 600\n'