Deployment

Secure baseline

Canonical policy posture — use with scenario labs before production claims.

Recommended deployment

Default install is tuned for observation and low false positives. High-risk AI engineering teams should adopt this baseline before claiming supply-chain or exfil protection.

See scenarios with default vs hardened
SettingDefaultBaseline
enforcement_mode

Without enforce, ES and file governors record would-block but do not deny.

monitor (log, allow AUTH)enforce
Observation mode (app)

Observation projects denies to alerts; use only while learning false positives.

Often on for new installsOff when tuning is complete
supply_chain_install_gate

Stops or defers package-manager AUTH before postinstall scripts run.

disabledenabled + npm, pnpm, pip, uv in intercept list
Behavioral rules (supply chain)

npm → node → osascript should not be alert-only in high-risk teams.

transitive_package_execution: alertPromote to block or pair with gate + HITL
blocked_domains

NE path blocks agentic egress to listed domains.

emptyOrg C2 blocklist + exfil destinations
AI installs

Writes supply-chain verdict + capability lineage for semantic contracts.

raw npm / pip from agentmf wrap npm install … where possible
Starter policy fragmentFull policy guide →
{
  "version": 1,
  "enforcement_mode": "enforce",
  "supply_chain_install_gate_enabled": true,
  "supply_chain_intercept_binaries": ["npm", "pnpm", "yarn", "pip", "pip3", "uv"],
  "supply_chain_defer_unknowns": true,
  "blocked_domains": ["your-org-blocklist.example"],
  "behavioral_network_block_threshold": 100
}