Resume ClawedBack

Restart the server and polling after a Claude Code session restart. Self-healing — if the config file is missing, it detects the setup and rebuilds it.

Step 1: Locate the project

PROJECT_ROOT="$(pwd)"
echo "Project root: $PROJECT_ROOT"
ls "$PROJECT_ROOT/.claude/skills/oc-poll/scripts/main.py" 2>/dev/null && echo "Server found" || echo "ERROR: Not in an ClawedBack project directory"

If .claude/skills/oc-poll/scripts/main.py doesn't exist, tell the user to cd into their ClawedBack directory first. STOP.

Step 2: Check for existing config

cat "$PROJECT_ROOT/data/clawedback.json" 2>/dev/null || echo "NO_CONFIG"

If config exists → skip to Step 5 (use the saved config).

If NO_CONFIG → proceed to Step 3 (auto-detect and rebuild).

Step 3: Auto-detect the setup

Run all of these to determine the environment:

# Who are we?
echo "USER=$(whoami)"
echo "UID=$(id -u)"

# Are we in tmux?
[ -n "$TMUX" ] && echo "TMUX=yes" || echo "TMUX=no"

# Is IS_SANDBOX set?
[ "$IS_SANDBOX" = "1" ] && echo "SANDBOX=yes" || echo "SANDBOX=no"

# Is there a venv?
[ -d "$PROJECT_ROOT/.venv" ] && echo "VENV=yes" || echo "VENV=no"

# What Python are we using?
which python3

# Can we find the venv python?
[ -f "$PROJECT_ROOT/.venv/bin/python" ] && echo "VENV_PYTHON=$PROJECT_ROOT/.venv/bin/python" || echo "VENV_PYTHON=none"

# Is FastAPI importable from system python?
python3 -c "import fastapi" 2>/dev/null && echo "SYSTEM_FASTAPI=yes" || echo "SYSTEM_FASTAPI=no"

# Is FastAPI importable from venv python?
[ -f "$PROJECT_ROOT/.venv/bin/python" ] && "$PROJECT_ROOT/.venv/bin/python" -c "import fastapi" 2>/dev/null && echo "VENV_FASTAPI=yes" || echo "VENV_FASTAPI=no"

# SSL certs?
[ -f "$PROJECT_ROOT/.claude/skills/oc-poll/scripts/fullchain.pem" ] && echo "SSL_LOCAL=yes" || echo "SSL_LOCAL=no"
ls /etc/letsencrypt/live/*/fullchain.pem 2>/dev/null | head -1 || echo "SSL_LETSENCRYPT=none"

# What port?
echo "PORT=${OC_PORT:-8080}"

# Host URL — check env, then detect adapter IP
echo "HOST_URL=${OC_HOST_URL:-}"
hostname -I 2>/dev/null | awk '{print "ADAPTER_IP=" $1}' || ip -4 addr show scope global | grep -oP '(?<=inet\s)\d+(\.\d+){3}' | head -1 | xargs -I{} echo "ADAPTER_IP={}"

# Whisper config?
echo "WHISPER_MODEL=${OC_WHISPER_MODEL:-turbo}"
echo "WHISPER_DEVICE=${OC_WHISPER_DEVICE:-auto}"

# Non-root users on the system
ls /home/ 2>/dev/null

# Is the server already running?
curl -sk http://localhost:${OC_PORT:-8080}/api/health 2>/dev/null || echo "SERVER_DOWN"

# Claude binary location
command -v claude 2>/dev/null || echo "$HOME/.local/bin/claude"

Step 4: Determine mode from detection

Use this logic:

if UID == 0:
    if IS_SANDBOX == 1:
        MODE = "c"    # System install, root, no venv, no permissions
    else:
        MODE = "b"    # Portable root with venv
else:
    MODE = "a"        # Portable user with venv

For the Python command:

if MODE == "c" (no venv):
    PYTHON = "python3"
    PIP_FLAGS = "--break-system-packages"
else:
    PYTHON = "$PROJECT_ROOT/.venv/bin/python"
    # Verify venv python works:
    $PYTHON -c "import fastapi" || PYTHON = "python3"  # fallback

For SSL, check in order:

1. Let's Encrypt certs at /etc/letsencrypt/live/*/ → use --public and --private with those paths
2. Local certs at .claude/skills/oc-poll/scripts/fullchain.pem and .claude/skills/oc-poll/scripts/privkey.pem → use --ssl
3. Neither → no SSL (HTTP)

For the non-root username (Mode B/C):

• Check ls /home/ for a single user → use that
• If multiple users, check data/sessions/install_mode.json for a saved username
• If still unknown, ask the user

For the claude binary:

CLAUDE_BIN="$(command -v claude 2>/dev/null || echo "$HOME/.local/bin/claude")"
# For root, also check the non-root user's path:
[ "$(id -u)" -eq 0 ] && [ -f "/home/$USERNAME/.local/bin/claude" ] && CLAUDE_BIN="/home/$USERNAME/.local/bin/claude"

Step 5: Start the server

First check if it's already running:

curl -sk http://localhost:${OC_PORT:-8080}/api/health 2>/dev/null

If health check succeeds → server is already running, skip to Step 6.

If not running, build the server command from the config (or detected values):

First kill any leftover server process:

pkill -f "python3 main.py" 2>/dev/null || true
sleep 1

Mode A (user, venv):

cd "$PROJECT_ROOT/.claude/skills/oc-poll/scripts" && source "$PROJECT_ROOT/.venv/bin/activate"
nohup python main.py $SSL_FLAGS > "$PROJECT_ROOT/data/logs/server.log" 2>&1 &
sleep 1 && pgrep -f "python3 main.py" | tail -1 > "$PROJECT_ROOT/data/server.pid"

Mode B (root, venv):

cd "$PROJECT_ROOT/.claude/skills/oc-poll/scripts" && source "$PROJECT_ROOT/.venv/bin/activate"
nohup python main.py $SSL_FLAGS > "$PROJECT_ROOT/data/logs/server.log" 2>&1 &
sleep 1 && pgrep -f "python3 main.py" | tail -1 > "$PROJECT_ROOT/data/server.pid"

Mode C (root, system python):

cd "$PROJECT_ROOT/.claude/skills/oc-poll/scripts"
nohup python3 main.py $SSL_FLAGS > "$PROJECT_ROOT/data/logs/server.log" 2>&1 &
sleep 1 && pgrep -f "python3 main.py" | tail -1 > "$PROJECT_ROOT/data/server.pid"

Where $SSL_FLAGS is:

• No SSL: (empty)
• Local SSL: --ssl
• Let's Encrypt: --public /etc/letsencrypt/live/$DOMAIN/fullchain.pem --private /etc/letsencrypt/live/$DOMAIN/privkey.pem

Verify it started:

sleep 3
kill -0 $(cat "$PROJECT_ROOT/data/server.pid") 2>/dev/null && echo "Process alive" || echo "PROCESS DIED"
curl -sk http://localhost:${OC_PORT:-8080}/api/health

If it died, show the log:

tail -30 "$PROJECT_ROOT/data/logs/server.log"

Help debug. STOP if server won't start.

Step 6: Start polling

CronCreate with cron "*/1 * * * *" and prompt "/oc-poll"

Verify with CronList.

Step 6.5: Recover interrupted workflows

Check for any in-progress approval chains that were interrupted by the crash:

cat "$PROJECT_ROOT/data/sessions/workflows.json" 2>/dev/null

If any workflow has "status": "in_progress", find the step with "status": "pending". Re-queue that step's approval request:

cd "$PROJECT_ROOT/.claude/skills/oc-poll/scripts" && python queue_manager.py write '{
  "content": "**Resumed workflow: WORKFLOW_NAME** (Step N/M)\n\nThis workflow was interrupted and is resuming.\n\nNext action: STEP_DESCRIPTION\n```\nSTEP_COMMAND\n```\n\nApprove? Reply **yes** or **no**.",
  "type": "approval_request",
  "metadata": {"workflow_id": "ID", "step": N, "total_steps": M}
}'

If no workflows.json exists or no workflows are in progress, skip this step.

Step 7: Run one poll cycle to confirm everything works

cd "$PROJECT_ROOT/.claude/skills/oc-poll/scripts" && $PYTHON queue_manager.py peek

If this returns valid JSON ({"pending": N}), the whole stack is working.

Step 8: Write config (only after everything works)

Only write the config file AFTER Steps 5-7 all succeeded. This way a broken config is never persisted.

Write data/clawedback.json:

cat > "$PROJECT_ROOT/data/clawedback.json" << JSONEOF
{
  "mode": "$MODE",
  "project_root": "$PROJECT_ROOT",
  "username": "$USERNAME",
  "host_url": "$HOST_URL",
  "python": "$PYTHON",
  "claude_bin": "$CLAUDE_BIN",
  "port": ${OC_PORT:-8080},
  "ssl": {
    "enabled": $SSL_ENABLED,
    "cert": "$SSL_CERT",
    "key": "$SSL_KEY"
  },
  "whisper": {
    "model": "${OC_WHISPER_MODEL:-turbo}",
    "device": "${OC_WHISPER_DEVICE:-auto}"
  },
  "server_cmd": "$SERVER_CMD",
  "resume_cmd": "$RESUME_CMD",
  "configured_at": "$(date -Iseconds)"
}
JSONEOF

Where:

• $HOST_URL is the adapter IP (auto-detected from hostname -I), or from OC_HOST_URL env var, or from the previous config. Falls back to adapter IP if unknown.
• $SSL_ENABLED is true or false
• $SSL_CERT / $SSL_KEY are the cert paths or empty strings
• $SERVER_CMD is the full command to start the server (for future reference)
• $RESUME_CMD is the full tmux command to restart ClawedBack:
  • Mode A: tmux new -s clawedback 'cd $PROJECT_ROOT && claude'
  • Mode B: tmux new -s clawedback 'cd $PROJECT_ROOT && sudo -E $CLAUDE_BIN'
  • Mode C: tmux new -s clawedback 'cd $PROJECT_ROOT && sudo -E $CLAUDE_BIN --dangerously-skip-permissions'

Step 9: Summary

ClawedBack resumed!

Mode: $MODE_NAME
Server: $PROTOCOL://localhost:$PORT (PID: $PID)
Polling: CronCreate every 1 minute
Auth token: (run /oc-token to view)

To reconnect to this tmux session later:
  tmux attach -t clawedback

If tmux wasn't detected, warn:

WARNING: You are not in tmux. Polling will stop when this terminal closes.
To fix: exit, then run:
  $RESUME_CMD

---

Using the Config (for other skills)

Any skill can read the config to find paths and settings:

cat "$PROJECT_ROOT/data/clawedback.json"

Or specific fields:

python3 -c "import json; c=json.load(open('$PROJECT_ROOT/data/clawedback.json')); print(c['python'])"