Hermes Backup Repo

Hermes Backup Repo — Disaster Recovery

WHAT

The Senna profile's entire state (config, skills, plugins, memory databases, key dotfiles) is backed up to a private GitHub repo: <your-github-username>/Hermes (cloned at ~/Hermes/).

Secrets (.env, auth.json) are never committed — the restore script prompts for them.

Two scripts in the repo root:

• backup.sh — snapshots current state into senna-profile/
• restore.sh — recovers from senna-profile/ onto a fresh machine

TRIGGER CONDITIONS

Use this skill when:

• User says "backup", "backup my Hermes", "backup repo", "backup the profile"
• User says "restore", "recover", "disaster recovery"
• User asks about the Hermes backup strategy or where their config is backed up
• User asks about secrets policy for the backup repo
• You've made significant changes to the profile (new skills, config changes) and want to snapshot
• User mentions crash/recovery/fresh install scenario

REPO STRUCTURE

~/Hermes/
├── README.md
├── backup.sh              # Snapshot script (run regularly)
├── restore.sh             # Recovery script (run on fresh machine)
├── senna-profile/
│   ├── .backup-manifest.txt    # Documents what's included/excluded
│   ├── config/
│   │   ├── config.yaml
│   │   ├── SOUL.md
│   │   ├── channel_directory.json
│   │   ├── .icarus-state.json
│   │   └── gateway_state.json
│   ├── skills/                 # Full copy of all profile skills
│   ├── plugins/                # plugin.yaml metadata only (not plugin code)
│   ├── memory/
│   │   ├── mnemosyne/data.tar.gz    # Compressed (~4 MB from 650 MB raw)
│   │   ├── ~~state.db.tar.gz~~      # EXCLUDED — 192 MB compressed, exceeds GitHub 100 MB limit
│   │   ├── lcm.db.tar.gz            # Compressed
│   │   ├── kanban.db.tar.gz
│   │   └── response_store.db.tar.gz
│   ├── home/                  # .zshrc, .gitconfig, gbrain_setup_status.json
│   ├── scripts/               # kanban-gate, security scanner, etc.
│   └── vault/                 # Obsidian-style vault notes

BACKUP PROCEDURE

cd ~/Hermes
./backup.sh                     # Collects state into senna-profile/
git add -A                      # Stage everything
git commit -m "backup $(date +%Y%m%d)"
git push

The backup script handles:

• Config files — direct copy from ~/.hermes/profiles/senna/
• Skills — rsync with delete (mirrors current state)
• Plugins — copies plugin.yaml metadata only
• Memory DBs — compressed via tar czf to stay under GitHub's 100 MB file limit
• Mnemosyne — compressed, excluding stale GGUF model files (~650 MB → ~4 MB)
• Home dotfiles — .zshrc, .gitconfig, gbrain status
• Scripts — security helpers, kanban gate
• Vault — any notes in the profile vault directory

WHAT'S EXCLUDED (never in repo)

SECRETS POLICY

Secrets are never committed to the repo. At restore time, the user is prompted:

1. Path to their .env file (or creates one manually)
2. Path to auth.json (or lets Hermes regenerate it)

RESTORE PROCEDURE (fresh machine)

# 1. Install Hermes Agent
# 2. Clone the backup repo
git clone https://github.com/<your-github-username>/Hermes.git

# 3. Run restore script
cd Hermes
./restore.sh [--profile senna] [--dry-run]

# 4. Follow prompts for secrets (.env + auth.json)
# 5. Start Hermes:  hermes
# 6. Verify skills: /skills

PATH NOTE

The shell's $HOME resolves to ~/.hermes/profiles/senna/home/ (not ~/). The backup script uses explicit absolute paths:

• HERMES_ROOT="~/.hermes"
• PROFILE_DIR="$HERMES_ROOT/profiles/senna"
• HOME_DIR="$PROFILE_DIR/home"

VERIFYING A BACKUP

After running backup.sh + git push, verify:

cd ~/Hermes && git log --oneline -1
# Should show: backup YYYYMMDD-HHMMSS

PITFALLS

• GitHub 100 MB file limit — SQLite DBs are compressed via tar czf before committing. If a DB grows past ~100 MB compressed, it won't push. In that case, split into chunks or use Git LFS.
• $HOME resolution — The shell's $HOME is ~/.hermes/profiles/senna/home/, so ~/ resolves through the profile symlink. Always use absolute paths.
• Old large files in git history — If an uncompressed DB was accidentally committed, force-push from clean history won't work if it's still in a prior commit. Use git reset --soft <clean-sha> to strip large blobs from history.
• Force push — Since this is a single-user repo, force-pushing to clean history is safe, but make sure your local state is complete first.
• state.db is the largest — At ~182 MB raw / ~78 MB compressed. This is the primary candidate for Git LFS if it grows further.