deploy_phase7.sh

#!/usr/bin/env bash
# deploy_phase7.sh — FCP v3 live cutover
#
# Run from the repo root on the Mac Mini:
#   bash scripts/deploy_phase7.sh [--dry-run]
#
# What it does (in order):
#   1. Preflight checks
#   2. Back up SOUL.md files and existing chains.jsonl (if any)
#   3. Deploy fcp_hook.py to ~/.hermes/fleet/hooks/
#   4. Deploy the FCP plugin per-profile (fcp-v3/plugin/ → each profile's plugins/fcp/)
#   5. Enable plugin in each profile's config.yaml (plugins.enabled: [fcp])
#   6. Patch all four SOUL.md files (FCP v2 → v3 stub)
#   7. Bootstrap Arch if its service is not loaded in launchd
#   8. Restart all four gateways (hard cutover — all at once)
#   9. Health check: confirm all four gateways are running
#
# Rollback: set FCP_HOOK_DISABLED=1 in each profile's .env and restart gateways.
# See IMPLEMENTATION.md §7 for the full rollback procedure.
#
# Flags:
#   --dry-run   Run checks and show diffs; no restarts, no file writes.

set -euo pipefail

# ─── config ────────────────────────────────────────────────────────────────

HERMES_HOME="$HOME/.hermes"
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
PLUGIN_SRC="$REPO_ROOT/fcp-v3/plugin"
HOOK_SRC="$REPO_ROOT/fcp-v3/hooks/fcp_hook.py"
HOOKS_DST="$HERMES_HOME/fleet/hooks"
BACKUP_DIR="$HERMES_HOME/fleet/backups/phase7_$(date +%Y%m%d_%H%M%S)"
AGENTS=(oracle arch eureka aurus)
LAUNCHD_DOMAIN="gui/$(id -u)"
PLIST_DIR="$HOME/Library/LaunchAgents"

DRY_RUN=0
if [[ "${1:-}" == "--dry-run" ]]; then
    DRY_RUN=1
    echo "=== DRY RUN MODE — no files will be written, no services restarted ==="
fi

# ─── helpers ───────────────────────────────────────────────────────────────

info()  { echo "  [INFO]  $*"; }
ok()    { echo "  [ OK ]  $*"; }
warn()  { echo "  [WARN]  $*"; }
fail()  { echo "  [FAIL]  $*" >&2; exit 1; }

run() {
    if [[ $DRY_RUN -eq 1 ]]; then
        echo "  [DRY]   would run: $*"
    else
        "$@"
    fi
}

# ─── step 1: preflight checks ──────────────────────────────────────────────

echo
echo "╔══════════════════════════════════════════╗"
echo "║  FCP v3 — Phase 7 Deploy  (preflight)   ║"
echo "╚══════════════════════════════════════════╝"
echo

# Plugin source files
for f in plugin.yaml __init__.py; do
    [[ -f "$PLUGIN_SRC/$f" ]] || fail "Missing plugin file: $PLUGIN_SRC/$f — is the repo up to date?"
done
ok "Plugin source files present ($PLUGIN_SRC)"

# fcp_hook.py
[[ -f "$HOOK_SRC" ]] || fail "fcp_hook.py not found at $HOOK_SRC"
ok "fcp_hook.py present"

# patch_soul.py
[[ -f "$REPO_ROOT/scripts/patch_soul.py" ]] || fail "patch_soul.py not found"
ok "patch_soul.py present"

# All four SOUL.md files
for agent in "${AGENTS[@]}"; do
    soul="$HERMES_HOME/profiles/$agent/SOUL.md"
    [[ -f "$soul" ]] || fail "Missing $soul"
done
ok "All four SOUL.md files found"

# python3
command -v python3 >/dev/null || fail "python3 not found in PATH"
ok "python3 available"

# yq or python — for config.yaml edits
if ! command -v yq &>/dev/null && ! python3 -c "import yaml" &>/dev/null; then
    warn "Neither yq nor python3 yaml module found — config.yaml plugin entry must be added manually"
    YAML_TOOL="manual"
else
    YAML_TOOL="python3"
    ok "yaml editing tool available"
fi

# ─── dry-run SOUL diff ──────────────────────────────────────────────────────

if [[ $DRY_RUN -eq 1 ]]; then
    echo
    echo "=== SOUL.md diff preview ==="
    python3 "$REPO_ROOT/scripts/patch_soul.py" --dry-run
    echo
    echo "=== plugin.yaml that will be installed ==="
    cat "$PLUGIN_SRC/plugin.yaml"
    echo
    echo "=== plugin install targets ==="
    for agent in "${AGENTS[@]}"; do
        echo "  $HERMES_HOME/profiles/$agent/plugins/fcp/"
    done
    echo
    echo "=== END DRY RUN — exiting without changes ==="
    exit 0
fi

# ─── step 2: backup ────────────────────────────────────────────────────────

echo
echo "── Step 2: backup ─────────────────────────────────────────────"
mkdir -p "$BACKUP_DIR"

for agent in "${AGENTS[@]}"; do
    src="$HERMES_HOME/profiles/$agent/SOUL.md"
    cp "$src" "$BACKUP_DIR/SOUL_${agent}.md"
    info "Backed up $agent SOUL.md"
done

# Backup existing chains.jsonl if present
chains="$HERMES_HOME/fleet/chains.jsonl"
[[ -f "$chains" ]] && { cp "$chains" "$BACKUP_DIR/chains.jsonl.bak"; info "Backed up chains.jsonl"; }

# Backup existing fcp_hook.py if present
existing_hook="$HOOKS_DST/fcp_hook.py"
[[ -f "$existing_hook" ]] && { cp "$existing_hook" "$BACKUP_DIR/fcp_hook.py.bak"; info "Backed up existing fcp_hook.py"; }

ok "Backups written to $BACKUP_DIR"

# ─── step 3: deploy fcp_hook.py ────────────────────────────────────────────

echo
echo "── Step 3: deploy fcp_hook.py ──────────────────────────────────"
run mkdir -p "$HOOKS_DST"
run cp "$HOOK_SRC" "$HOOKS_DST/fcp_hook.py"
ok "fcp_hook.py deployed to $HOOKS_DST/"

# ─── step 4: deploy plugin per-profile ─────────────────────────────────────

echo
echo "── Step 4: deploy plugin per-profile ───────────────────────────"
for agent in "${AGENTS[@]}"; do
    dst="$HERMES_HOME/profiles/$agent/plugins/fcp"
    run mkdir -p "$dst"
    run cp "$PLUGIN_SRC/plugin.yaml" "$dst/plugin.yaml"
    run cp "$PLUGIN_SRC/__init__.py" "$dst/__init__.py"
    info "Plugin installed to $dst/"
done
ok "Plugin deployed to all four profiles"

# ─── step 5: enable plugin in config.yaml ──────────────────────────────────

echo
echo "── Step 5: enable plugin in config.yaml ────────────────────────"
for agent in "${AGENTS[@]}"; do
    cfg="$HERMES_HOME/profiles/$agent/config.yaml"
    if [[ ! -f "$cfg" ]]; then
        warn "config.yaml not found for $agent — skipping plugin enable"
        continue
    fi

    if grep -q "fcp" "$cfg" 2>/dev/null; then
        info "$agent: fcp already present in config.yaml"
        continue
    fi

    if [[ "$YAML_TOOL" == "python3" ]]; then
        # Append plugins.enabled block, or add fcp to existing list
        python3 - "$cfg" <<'PYEOF'
import sys, re

path = sys.argv[1]
text = open(path).read()

# If plugins.enabled already exists, inject fcp
if re.search(r'^plugins:\s*\n.*enabled:', text, re.MULTILINE | re.DOTALL):
    # Find the enabled list and append fcp if not present
    text = re.sub(
        r'(plugins:\s*\n\s*enabled:\s*\n)((?:\s+-\s+\S+\n)*)',
        r'\1\2    - fcp\n',
        text
    )
elif 'plugins:' in text:
    # plugins: block exists but no enabled key
    text = re.sub(r'(plugins:\s*\n)', r'\1  enabled:\n    - fcp\n', text)
else:
    # No plugins block at all — append to end
    text = text.rstrip('\n') + '\n\nplugins:\n  enabled:\n    - fcp\n'

open(path, 'w').write(text)
print(f"  [INFO]  {path}: plugins.enabled += fcp")
PYEOF
    else
        warn "$agent: add 'plugins:\\n  enabled:\\n    - fcp' to $cfg manually"
    fi
done
ok "Plugin enabled in config.yaml for all four profiles"

# ─── step 6: patch SOUL.md files ───────────────────────────────────────────

echo
echo "── Step 6: patch SOUL.md files ─────────────────────────────────"
python3 "$REPO_ROOT/scripts/patch_soul.py"
ok "SOUL.md files patched"

# ─── step 7: bootstrap Arch if needed ──────────────────────────────────────

echo
echo "── Step 7: Arch gateway ────────────────────────────────────────"
ARCH_PLIST="$PLIST_DIR/ai.hermes.gateway-arch.plist"
if ! launchctl print "$LAUNCHD_DOMAIN/ai.hermes.gateway-arch" &>/dev/null; then
    if [[ -f "$ARCH_PLIST" ]]; then
        info "Arch service not loaded — bootstrapping"
        run launchctl bootstrap "$LAUNCHD_DOMAIN" "$ARCH_PLIST"
        ok "Arch bootstrapped"
    else
        warn "Arch plist not found at $ARCH_PLIST"
        warn "Run 'hermes -p arch gateway install' first, then re-run this script"
    fi
else
    ok "Arch service already in launchd domain"
fi

# ─── step 8: hard cutover — restart all four gateways ──────────────────────

echo
echo "── Step 8: hard cutover ────────────────────────────────────────"
info "Stopping all four gateways..."
for agent in "${AGENTS[@]}"; do
    label="ai.hermes.gateway-$agent"
    if launchctl print "$LAUNCHD_DOMAIN/$label" &>/dev/null; then
        run launchctl stop "$LAUNCHD_DOMAIN/$label" || warn "Stop $label returned non-zero"
    fi
done

sleep 2

info "Starting all four gateways..."
for agent in "${AGENTS[@]}"; do
    label="ai.hermes.gateway-$agent"
    run launchctl kickstart -k "$LAUNCHD_DOMAIN/$label" || {
        warn "kickstart $label failed — check $HERMES_HOME/profiles/$agent/logs/gateway.error.log"
    }
done
ok "All four gateways kicked"

# ─── step 9: health check ──────────────────────────────────────────────────

echo
echo "── Step 9: health check (15s settle) ───────────────────────────"
sleep 15

PASS_COUNT=0
FAIL_COUNT=0
for agent in "${AGENTS[@]}"; do
    label="ai.hermes.gateway-$agent"
    pid=$(launchctl print "$LAUNCHD_DOMAIN/$label" 2>/dev/null | awk '/pid =/ {print $3}')
    if [[ -n "$pid" && "$pid" != "0" ]]; then
        ok "$agent gateway running (PID $pid)"
        PASS_COUNT=$((PASS_COUNT + 1))
    else
        warn "$agent gateway NOT running — check $HERMES_HOME/profiles/$agent/logs/gateway.error.log"
        FAIL_COUNT=$((FAIL_COUNT + 1))
    fi
done

echo
echo "╔══════════════════════════════════════════╗"
if [[ $FAIL_COUNT -eq 0 ]]; then
    echo "║  ✓  All 4 gateways up. FCP v3 is live. ║"
    echo "╚══════════════════════════════════════════╝"
    echo
    echo "Next: run Phase A regression tests (IMPLEMENTATION.md §7.4):"
    echo "  python3 scripts/regression_phase7.py"
    echo
    echo "Rollback if needed:"
    echo "  Add FCP_HOOK_DISABLED=1 to each profile's .env, restart gateways."
    echo "  SOUL.md originals: $BACKUP_DIR"
else
    echo "║  ✗  $FAIL_COUNT gateway(s) failed to start.          ║"
    echo "╚══════════════════════════════════════════╝"
    echo
    echo "Check logs:"
    for agent in "${AGENTS[@]}"; do
        echo "  tail -50 $HERMES_HOME/profiles/$agent/logs/gateway.error.log"
    done
    echo
    echo "Originals backed up in: $BACKUP_DIR"
    echo "To rollback plugin: rm -rf \$HERMES_HOME/profiles/<agent>/plugins/fcp for each agent"
    exit 1
fi