加密交易代理

git clone https://github.com/algorytma/hermes-senpi-railway-template
cd hermes-senpi-railway-template

railway login
railway init

railway up

Container (hermes-agent:v0.6.0)
    │
    └─▶ bootstrap/prestart.sh
            ├── init-volume.sh          # first-run /data setup
            ├── run-migrations.sh       # schema version check
            ├── validate-risk.sh        # execution: risk_policy.yaml required
            └── scripts/render-config.py
                    │
                    ├── providers/provider_registry.yaml  ─┐
                    ├── providers/model_aliases.yaml        ├─▶ config.generated.yaml
                    ├── mcp/mcp_registry.yaml (filtered)  ─┘
                    └── mcp/mcp.generated.yaml
                                │
                          hermes runtime

Env vars (secrets)
    +
Canonical YAML → bootstrap/render-config.py → Generated config → Hermes runtime
  (you edit)                                    (auto-produced)    (reads this)

mode: copilot                           # human approval required for all actions
execution:
  dry_run_default: true                 # everything is dry-run until you change this
  allow_open_new_position: false        # cannot open new positions by default
symbols:
  allowlist: ["BTC", "ETH", "SOL"]
  default_action: deny                  # all non-allowlist symbols rejected
trading:
  max_leverage: 3
  max_position_size_usd: 300
  max_daily_loss_usd: 100

railway run bash scripts/install-skill.sh /data/skills/manifests/wolf-strategy.yaml

railway run bash scripts/backup-export.sh
# Output: /data/backups/exports/backup-YYYYMMDD-HHMMSS.tar.gz

railway run bash scripts/export-config.sh

railway run bash scripts/backup-restore.sh /data/backups/exports/backup-20260404-000000.tar.gz
railway redeploy

cp .env.example .env
# Edit .env — add at minimum OPENROUTER_API_KEY

docker build --build-arg HERMES_VERSION=v0.6.0 -t hermes-senpi-trader .

docker run --rm \
  -e ACTIVE_PROFILE=analysis \
  -e DATA_DIR=/data \
  -e OPENROUTER_API_KEY=your-key \
  -v $(pwd)/.tmpdata:/data \
  hermes-senpi-trader

# Check logs — bootstrap should complete and Hermes should start

docker run --rm ... bash scripts/healthcheck.sh

# 1. Take a backup
railway run bash scripts/backup-export.sh

# 2. Update HERMES_VERSION in railway.toml and redeploy
# railway.toml → [build.args] → HERMES_VERSION = "v0.7.0"

railway up

# 3. Bootstrap auto-runs migrations and regenerates config

bash tests/smoke_render.sh
bash tests/smoke_backup.sh
bash tests/validate_risk_test.sh

hermes-senpi-trader/
├── Dockerfile                      # hermes-agent:v0.6.0 based, pinned
├── railway.toml                    # Railway deploy config
├── .env.example                    # Environment variable template
├── CLAUDE.md                       # AI assistant guidance for this repo
├── SECURITY.md                     # Security model and operator checklist
├── LICENSE
│
├── bootstrap/                      # Container lifecycle scripts
│   ├── prestart.sh                 # Main entrypoint (6-step boot)
│   ├── init-volume.sh              # First-run /data initialization
│   ├── validate-risk.sh            # Execution profile risk gate
│   └── run-migrations.sh           # Schema version migrations
│
├── scripts/                        # Utility scripts
│   ├── render-config.py            # Canonical YAML → Hermes config (no keys on disk)
│   ├── install-skill.sh            # Pinned-commit skill installer (SHA verified)
│   ├── backup-export.sh            # Full volume backup (sessions excluded)
│   ├── backup-restore.sh           # Restore from backup (path traversal guard)
│   ├── export-config.sh            # Config-only export
│   ├── healthcheck.sh              # Container health validation
│   └── lib/
│       └── redact.sh               # Secret redaction helper library
│
├── config/                         # Canonical config templates (.example.yaml)
│   ├── providers/
│   │   ├── provider_registry.example.yaml
│   │   └── model_aliases.example.yaml
│   ├── mcp_registry.example.yaml
│   ├── profiles.example.yaml
│   ├── risk_policy.example.yaml
│   ├── symbol_policy.example.yaml
│   └── skills/
│       ├── wolf-strategy.example.yaml
│       ├── senpi-market-scan.example.yaml
│       └── senpi-stop-manager.example.yaml
│
├── defaults/workspace/             # First-run agent workspace content
│   ├── AGENTS.md
│   ├── BOOTSTRAP.md
│   └── USER.md
│
├── templates/
│   └── hermes-config.j2            # Jinja2 template for config generation
│
├── docker/
│   └── entrypoint.sh               # Docker SIGTERM handler
│
├── migrations/
│   └── 0001_initial_schema.sh
│
├── tests/
│   ├── smoke_render.sh             # Config render smoke test (API key leak check)
│   ├── smoke_backup.sh             # Backup/restore round-trip test
│   └── validate_risk_test.sh       # Risk policy unit tests (7 cases)
│
├── .github/workflows/
│   └── ci.yml                      # CI: shellcheck, flake8, smoke tests
│
└── docs/
    ├── architecture.md
    ├── deployment.md
    ├── backup-restore.md
    ├── migration.md
    └── security.md

[ ] Volume mounted at /data (minimum 5 GB)
[ ] ACTIVE_PROFILE=analysis (NOT execution for first deploy)
[ ] OPENROUTER_API_KEY set in Railway Variables (not in any config file)
[ ] DATA_DIR=/data in Railway Variables
[ ] HERMES_VERSION pinned (e.g., v0.6.0) — never use 'latest'
[ ] .env file is in .gitignore and not committed to git
[ ] railway.toml contains no secret values
[ ] CI green: shellcheck, python lint, render smoke, risk tests

[ ] At least 7 days of analysis profile without errors
[ ] risk_policy.yaml reviewed — all 8 required fields intentionally set
[ ] dry_run_default: true (YAML boolean, not string)
[ ] allow_open_new_position: false
[ ] mode: copilot
[ ] symbols.allowlist has only intended symbols
[ ] symbols.default_action: deny
[ ] max_leverage <= 3 for initial testing
[ ] max_daily_loss_usd set to acceptable ceiling
[ ] Backup taken before switching profiles
[ ] SENPI_AUTH_TOKEN valid
[ ] ACTIVE_PROFILE=execution → redeploy → logs show "Risk policy validation passed"
[ ] Healthcheck passes: bash scripts/healthcheck.sh

[ ] Paper trading tested for at least 2 weeks
[ ] Paper trade outcomes reviewed, strategy validated
[ ] max_position_size_usd <= 5% of portfolio per trade
[ ] max_daily_loss_usd < 3% of total capital
[ ] max_leverage <= 3 (only increase after proven track record)
[ ] Telegram or alert channel configured and tested
[ ] Backup taken immediately before going live
[ ] Rollback plan documented: Railway Deployments -> Rollback
[ ] Someone monitors the first 24h of live trading
[ ] dry_run_default: false — set explicitly and intentionally
[ ] allow_open_new_position: true — set explicitly and intentionally
[ ] Mode stays copilot or semi-auto — never jump directly to auto
[ ] Bootstrap logs reviewed after deploy: validate-risk passed

[ ] bash scripts/backup-restore.sh <archive.tar.gz> --yes
[ ] Railway redeploy
[ ] Bootstrap logs: "Config generation completed"
[ ] Healthcheck passes: bash scripts/healthcheck.sh
[ ] /data/providers/provider_registry.yaml exists and valid
[ ] /data/risk/risk_policy.yaml exists
[ ] /data/workspace/AGENTS.md exists
[ ] Generated configs present: ls /data/.hermes/*/config.generated.yaml
[ ] Sessions cleared: ls /data/.hermes/*/sessions/ (should be empty)
[ ] API keys valid on new host: python3 scripts/render-config.py --validate-only
[ ] ACTIVE_PROFILE=analysis tested before enabling execution

[ ] Backup taken: bash scripts/backup-export.sh
[ ] Backup downloaded and verified locally: tar -tzf backup.tar.gz
[ ] New HERMES_VERSION changelog reviewed for breaking changes
[ ] HERMES_VERSION pinned in railway.toml (never 'latest')
[ ] CI green before deploy
[ ] Bootstrap migration logs reviewed after redeploy
[ ] Healthcheck passes on new version
[ ] Analysis profile tested before enabling execution