AAgentProof

Sample-only walkthrough — no real workspace data

What the deeper workspace experience looks like

A founder/tester-safe preview using sample data only. Lets you judge the deeper trial / workspace experience before sign-in is configured.

Sample onlyNo real customer dataPublic-safe

Sample agent

Sample: Order-routing assistant (preview only)

Action-taking agent in a Microsoft-style environment

Zone: action takingPack v2 (sample)

Readiness score

64

medium band

Top recommended controls

In a real workspace, these are wired to your improvement cycle. In this preview, they show the shape of the recommendation engine.

action authority

Action authority

Next action: Confirm idempotency on every write trigger; document a tested rollback.

human oversight

Human oversight

Next action: Add a named approver role for any action that changes customer-visible records.

testing and evidence

Testing and evidence

Next action: Include a red-team pass against the top 20 anonymised real-world prompts.

Evidence expectations

  • Action allow-list snapshot (current)
  • Per-action audit trace (sanitised, no PII)
  • Tested rollback procedure record
  • Reviewer roster + sample approval log

Report-linked guidance

Why this score?

Three control families are at Managed; one is at Basic. Closing the action-authority gap takes the agent to Strong.

What to fix first

Idempotency on the order-routing write action — the cheapest control against duplicate downstream effects.

What is preserved

Reports are bound to the intelligence pack version that produced them. Reassessment is recommended when the pack version changes.

Improvement-cycle suggestions

  1. Run a rollback rehearsal and store the record as evidence.
  2. Add a named approver for sensitive write actions.
  3. Rerun the agent against the top 20 anonymised real-world prompts.

Radar relevance for this agent

  • Connector scope changes — watching
  • Audit minimum-evidence expectations — approved
Open the AI landscape radar →

Related Learn entries (deep links)

In a real workspace, every finding deep-links into the Library entry that produced it.

What the real workspace unlocks

Beyond this preview

  • Full per-zone control checklists
  • Report-linked deep dives into each finding
  • Improvement-cycle suggestions wired into your workspace
  • Stale-source warnings on affected reports
  • Reassessment trigger when pack version changes
Start the free assessment →

Ready to assess your own agent?

Start the free assessment to apply this guidance to a real agent. No payment. No public registration required.

A few honest things about AgentProof

  • · AgentProof is a readiness assessment, not an official audit.
  • · Every recommendation cites the intelligence pack version it came from.
  • · Intelligence updates go through a human review gate.
  • · AgentProof does not speak on behalf of Microsoft or any vendor.