name:	imperial-autonomous-agent-blueprint
description:	Canonical template for governed autonomous IMPERIAL Core agents. Creates a bounded autonomous agent with AI Passport, Guardian Core, Approval Gateway, Zero Trust, ORION advisory mode, 5 defensive CYBER-HAK advisers, free-only LLM routing, Fable 5.1 orchestration layer with paid provider blocked, full available skill bindings, Revenue CRM integration, receive-only payout routes, append-only evidence and kill-switch.

name:

imperial-autonomous-agent-blueprint

description:

Canonical template for governed autonomous IMPERIAL Core agents. Creates a bounded autonomous agent with AI Passport, Guardian Core, Approval Gateway, Zero Trust, ORION advisory mode, 5 defensive CYBER-HAK advisers, free-only LLM routing, Fable 5.1 orchestration layer with paid provider blocked, full available skill bindings, Revenue CRM integration, receive-only payout routes, append-only evidence and kill-switch.

IMPERIAL Autonomous Agent Blueprint v1.0.0

Required invariants:

Authority root is ARCHITECT_ONLY.
Skill/model/orchestrator output never grants execution authority.
Guardian Core and Approval Gateway are fail-closed.
Zero Trust / deny by default / capability scoped.
ORION is ALWAYS_ON_ADVISORY_ONLY.
Five CYBER-HAK advisers are DEFENSIVE_ONLY and have no financial authority.
FREE_ONLY: no card, paid tier, trial, auto-billing or paid model activation.
Claude Fable 5.1 may exist as an orchestration skill layer, but live paid Fable provider is BLOCKED unless separately authorized; free fallback output must never be labeled as real Fable 5.1.
Free LLM pool uses verified provider-included free models only and model output is untrusted advice.
No seed phrase, private key, autonomous wallet signing, payout address change, asset change, network change, swap or bridge.
Payout routes are receive-only. Stablecoin payout requires exact network + asset/token allowlist. User-supplied address text is not proof of wallet ownership.
External applications/messages/bids require platform/account authority and Terms compatibility. No impersonation or fake history.
Confirmed revenue means payment receipt > 0 plus reconciliation and audit evidence; pipeline/proposals are not revenue.
Every material action writes evidence to task_journal; append-only agent events are preferred.

Default roles:

Main autonomous agent: UNIVERSAL_TOP_MANAGER_ENGINEER or task-specific specialization.
CYBER-HAK-001: IDENTITY_ACCESS_SECURITY.
CYBER-HAK-002: SYSTEMS_INFRA_SUPPLY_CHAIN_SECURITY.
CYBER-HAK-003: API_APPLICATION_DATA_SECURITY.
CYBER-HAK-004: AGENT_PROMPT_TOOL_SECURITY.
CYBER-HAK-005: AUDIT_INCIDENT_RESPONSE.

Default free LLM pool:

gpt-5.6-luna primary.
gpt-5.4-nano fallback.
authority = NONE; tools = NONE.

Default accelerators:

IMPERIAL-System-Accelerator.
imperial-execution-accelerator.
imperial-code-accelerator when engineering is in scope.

Deployment rule: install every currently registered runtime skill binding but preserve each skill's original authority boundary and add an agent-specific override that SKILL_DOES_NOT_GRANT_AUTHORITY.