RobotnixAcademy

Week 12 · lesson

Investigation: Coordination Needs Contracts and Authority

system flow

From Agent Message to a Bounded Action Decision

  1. Agent Capability

    Declare what each component can do and its least-privilege boundary.

  2. Message Envelope

    Preserve sender, recipient, task, evidence, schema, and status.

  3. Schema Check

    Reject ambiguous or incompatible field meanings.

  4. Evidence Check

    Keep provenance and uncertainty attached through handoffs.

  5. Authority Gate

    Require explicit owner approval before external action.

  6. Accountable Outcome

    Record hold, release, failure, owner, and recovery path.

Read this concept flow as plain text
  1. Agent Capability. Declare what each component can do and its least-privilege boundary. LAB: 3 TRACES.
  2. Message Envelope. Preserve sender, recipient, task, evidence, schema, and status. CONTRACT: REQUIRED FIELDS.
  3. Schema Check. Reject ambiguous or incompatible field meanings. CHECK: ISO DATE.
  4. Evidence Check. Keep provenance and uncertainty attached through handoffs. TRANSPARENCY: SOURCE REF.
  5. Authority Gate. Require explicit owner approval before external action. RISK: FAIL CLOSED.
  6. Accountable Outcome. Record hold, release, failure, owner, and recovery path. ACTION: AUDIT RESULT.

An agent perceives an environment and takes actions. A multi-agent system adds handoffs: one agent's output becomes another agent's input. Complex behavior can emerge from those interactions, but so can ambiguity, duplicated action, missing evidence, and authority escalation.

A useful message envelope makes the handoff inspectable:

sender
recipient
task
schema version
evidence reference
requested action
authority status
handoff status

Completed approval-gate example

A planning agent proposes a fictional booking. The action agent can execute bookings, but the envelope states ownerApproved: false.

requested action: book
capability declared: yes
evidence attached: yes
schema valid: yes
owner approved: no
result: HOLD — no external action

Capability answers “can this component perform the operation?” Authority answers “may it perform this operation now?” The second does not follow from the first.

Responsible AI is a system obligation

Reliability and safety require failure tests and recovery. Privacy and security limit data and access. Transparency exposes AI use, evidence, and uncertainty. Accountability names who owns decisions and remediation. Fairness and inclusiveness require affected-group evidence and accessible participation, not a generic checkbox. A clean handoff cannot prove every principle, but a missing contract makes responsible operation harder to inspect.

Investigation

  1. Write the envelope for the completed example.
  2. Replace an ISO date with ambiguous 03/04; identify the failed contract.
  3. Trace what should happen when approval is absent.
  4. Name the accountable human role for release and incident response.
  5. Repair: “The agents agreed, so the system can act responsibly.”

Vocabulary lab

Flip the idea, not just the card

Explain the term before you reveal the back. Then compare your explanation with the definition, example, and warning.

1 / 5
Read all terms without animation
agent
A component that perceives inputs and selects actions within an environment. Example: A planning agent proposes a next step. Do not confuse it with: Agent capability does not imply unrestricted authority.
protocol
Declared rules for message exchange and interaction. Example: Every handoff must include status and evidence reference. Do not confuse it with: A protocol does not guarantee truthful content.
shared schema
A common structure and meaning for exchanged fields. Example: Dates must use YYYY-MM-DD. Do not confuse it with: Matching field names alone do not guarantee matching semantics.
handoff
The transfer of a task, result, or responsibility between components. Example: Research evidence passes to a planner. Do not confuse it with: A handoff should not erase provenance or ownership.
human authorization
Explicit permission from the accountable person required before a bounded action. Example: Owner approval is required before booking. Do not confuse it with: Human review is not implied by agent consensus.