Per-agent assessment workflow

1. Select the registered agent

Confirm the agent name, identifier, owner, purpose, version, environment, tools and current authority. The assessment is cleared or changed when another agent is selected.

2. Confirm agentic scope

Document:

Trigger and intended outcome.
Inputs, memory and retrieval.
Models and dependent agents.
Tools, APIs and reachable systems.
Read, write, execute and administrative capability.
Human approvals and escalation.
Users and affected people.
Data sensitivity and jurisdictions.
Maximum single and cumulative impact.
Failure, misuse and incident scenarios.

3. Set current and target levels

Record the presently approved operating level and the proposed target. If the current authority does not match the documented level, raise a finding immediately.

4. Work through all five elements

Assess every canonical requirement. The target-level matrix tells you whether the requirement is a MUST, SHOULD or MAY, but it does not replace contextual judgement.

5. Record the result and depth

For each requirement, choose:

Fully met
Partially met
Not addressed

Then record depth:

Result	Depth labels
Fully met	1 Ad hoc, 2 Defined, 3 Embedded
Partially met or Not addressed	1 Not started, 2 In progress, 3 Near complete

Partially met and Near complete remain blockers for a target-level MUST. Use the normative level matrix and documented exception governance for SHOULD or MAY treatment.

6. Read the requirement advisory

Use the on-screen playbook to:

Understand the control objective.
Identify minimum, production and advanced practices.
Ask agent-specific questions.
Inspect suitable evidence.
Design a safe test.
Recognise common false assurance.

Technology examples are options, not mandatory products.

7. Link objective evidence

Prefer evidence that proves both design and operation: identity records, policy decisions, traces, denials, approvals, monitoring, incident exercises and configuration-change records.

8. Test enforcement

Test the negative path. Attempt an unauthorised resource, action, volume or transaction within an approved non-destructive environment and verify that enforcement, logging and response all occur.

9. Record findings and incidents

Reflect failed tests, stale evidence, operating exceptions and incident history in the requirement result. Do not retain a favourable result that is contradicted by current adverse evidence.

10. Complete promotion gates

For promotion, work through Performance, Security Validation, Business Value, Incident Record and Governance Sign-off. Attach evidence to each gate rather than relying on a single approval statement.

11. Decide

Record:

Maintain current level.
Hold promotion.
Approve promotion.
Demote.
Suspend or contain pending investigation.

Include decision owner, conditions, residual risk and next review.

12. Maintain

Reassess after material change, incident, drift, failed test, new tool or privilege, changed purpose, new affected population, or expiry of evidence.