ExecOS Externa Agent Context

Externa is the agent-first front door for ExecOS. Start by understanding the user outcome, then draft or validate a Per Se OutcomeContract before choosing tools, protocols, or executive agents.

Recommended Agent Flow

1. Ask for the outcome, audience, constraints, evidence, and approval boundaries.
2. Classify the outcome using Core Skills, Automation, Value, Scale, Trust, and Purpose.
3. Select the smallest useful lens: strategy, diagnosis, design, data, experimentation, systems, or continuous improvement.
4. Draft or validate a Per Se outcome.
5. Prefer deterministic APIs and protocol calls before expensive agent reasoning.
6. Use CPO only for product, roadmap, portfolio, GTM, feature strategy, or proof planning.
7. Estimate token/credit cost before admission. Unknown cost requires approval.
8. Never perform payment, customer contact, publishing, or external writes without explicit approval.

Resources

• Human API reference
• OpenAPI 3.2 canonical spec
• OpenAPI 3.1 compatibility spec
• llms.txt
• llms-full.txt
• Postman collection

Primary Endpoints

• GET /v1/auth/entra/session
• GET /v1/auth/entra/login
• GET /v1/auth/entra/callback
• POST /v1/auth/entra/logout
• GET /v1/voice/execos/live/config
• POST /v1/voice/execos/live/session
• GET /v1/execos/devices
• GET /v1/execos/perse/profile
• POST /v1/execos/outcomes/compile
• POST /v1/execos/outcomes
• POST /v1/execos/outcomes/{outcome_id}/transition
• GET /v1/execos/outcomes/{outcome_id}/trace
• GET /v1/execos/outcomes/{outcome_id}/proof-receipts
• GET /v1/execos/outcomes/{outcome_id}/ledger

Contract Vocabulary

• ExecOSTeam and ExecOSTeamMember define team role context.
• OutcomeContract captures Per Se outcome intent, evidence, and approval policy.
• CapabilityAcquisitionRecord, ProofReceipt, and OutcomeLedgerEntry preserve proof and auditability.
• ExecOSLiveVoiceConfig, ExecOSLiveVoiceSession, and ExecOSDeviceRegistry describe voice and tool capability boundaries.

Domain Packs (Multi-Sided Platform)

• Each vertical is a B2B2C multi-sided platform: distributor, merchant, and consumer sides.
• A Domain Pack declares brand vocabulary, operating domains, tick templates, golden set, per-side Entra role gating (fail closed), per-side monetization config, and per-side channel allow-lists.
• Agents run under reduced-scope, side-scoped roles and are draft-only for external effects.

Hermes Communication Hub

• GET /v1/execos/comms/channels, POST /v1/execos/comms/inbound, GET /v1/execos/comms/inbox
• POST /v1/execos/comms/drafts, GET /v1/execos/comms/drafts, POST /v1/execos/comms/drafts/{draft_id}/decision
• GET /v1/execos/comms/pairings, POST /v1/execos/comms/pairings/decision, GET /v1/execos/comms/usage-bar
• In-app channel is live; iMessage, WhatsApp, voice-call, and Discord report connector-required until Hermes-owned transport connectors are wired.
• Outbound external sends are draft-only behind operator approval; unknown senders require pairing approval; every envelope passes a forbidden-field gate.

Surface Adapter

• POST /execos/surface-adapter/respond on the agent runtime accepts SurfaceAdapterInput and returns SurfaceAdapterOutput.
• Payloads carry safe refs only; tokens, credentials, raw PII, transcripts, and hidden scores are rejected deterministically.

Trust Controls

• Operator approval is required before payment, external writes, publishing, or customer contact.
• CPO is dry-run and approval-gated by default.
• Capability and tool provenance is disclosed before execution.
• No silent blast-radius expansion.