# Prime Bootstrap v0 ## 1) Environment ```bash source /srv/prime-core/scripts/prime-env.sh export PRIME_VAULT_KEY='set-a-strong-random-value' ``` ## 2) Validate ```bash prime status prime dispatch planning "design backup policy" prime dispatch coding "implement hello endpoint" ``` ## 3) Backup/Restore ```bash prime backup-now prime restore ``` ## 4) OAuth broker stubs Current `prime_core/oauth.py` is a stub interface (start/callback) and encrypted token vault. Replace `auth.example` URLs and token exchange with provider-specific OAuth implementations. ## 5) Queue worker + approvals ```bash prime queue add planning "design X" prime queue run-once prime queue list prime queue approve prime queue deny --reason "optional reason" ``` Worker service: ```bash sudo systemctl status prime-worker.service ``` ## 6) Telegram notifications Set in local runtime secrets (`.secrets/runtime.env`): - `TELEGRAM_BOT_TOKEN` - `TELEGRAM_CHAT_ID` Worker sends alerts for: - blocked approval tasks - failed tasks ## 7) Telegram control plane (beyond request/response) Prime now includes: - `prime-telegram-control.service` (inbound command channel) - `prime-autonomy-loop.service` (goal-driven autonomous queue seeding) Telegram commands: - `status` - `queue list` - `queue add ` - `queue approve ` - `queue deny [reason]` - `run once` - `mode autonomous on|off` - `goals show` - `goals add ` ## 8) Stability Sprint v1 Added: - Risk classification tiers (`prime_core/risk.py`) - Audit trail (`logs/audit.jsonl`) - Safer restore: - `prime restore --dry-run ` - preflight + rollback on failure - Telegram control guardrails for high-risk direct commands ## 9) Backup classes + verification (Stability Sprint v2) Prime now separates backups into classes: - `scripts/backup/code-backup.sh` -> git code/config backup - `scripts/backup/state-backup.sh` -> encrypted state snapshot (`task-queue.json`, `tokens.enc`) - `scripts/backup/secrets-backup.sh` -> encrypted secret escrow (optional, requires `PRIME_SECRET_BACKUP_PASSPHRASE`) Timers: - `prime-state-backup.timer` (15m) - `prime-restore-test.timer` (nightly dry-run restore) - `prime-audit-watch.timer` (15m anomaly check) ## 11) Telegram allowlist controls Prime responds only to allowlisted Telegram users. Owner (configured `TELEGRAM_CHAT_ID`) can manage list: - `allowlist show` - `allowlist add ` - `allowlist remove ` Owner cannot be removed from allowlist. ## 12) Prime proactive check-in loop Service: `prime-checkin-loop.service` Behavior: - Sends proactive check-ins every ~3 hours - Asks for goal/task direction when queue is clear - Quiet hours enforced: 11:00 PM–6:00 AM America/New_York - Personality: calm, strategic, concise (`prime_core/personality.py`) ## 13) Chat Mode v1 (conversational control) Prime can now interpret natural-language messages and not just strict commands. New controls: - `mode chat on|off` - `mode strict on|off` Behavior: - Interprets intent via OpenAI/Codex (`prime_core/chat_mode.py`) - For action requests: proposes plan and asks for explicit approval (`approve`) when strict/confirm needed - Keeps pending user action context in `state/pending-actions.json` ## 14) Chat Mode v1.2 (assistant-first behavior) - Owner fast-lane for conversational planning/research requests - Hard blocks only for tier4 destructive/external asks - Softer plan/confirm messaging for normal requests ## 15) Startup rehydrate protocol (Hot/Cold memory) - `STARTUP.md`, `HOT_MEMORY.md`, `COLD_INDEX.md` - generated `state/boot-pack.json` for compact startup context - timer: `prime-boot-pack.timer` (15m) - Telegram: `startup show`