Self-guided worksheet. This reading edition provides questions and evidence criteria. It does not calculate a score, produce a radar chart, compare peers, save answers, email results or generate a PDF.

Assess whether the team can explain what an agent may do, who owns its decisions and what happens when the work diverges from policy.

How to use it

Choose one workflow. For each question, record the answer, source, owner, date, gap and next check in your team’s document. Use “unknown” when the evidence is absent. Repeat after a material workflow change to compare your own records.

Define the inventory boundary

Which agents, connectors, models and delegated accounts are included?

Evidence to collect: An inventory with owners, data classes, permissions and explicit coverage gaps.

Make policy executable

What written expectation applies to each consequential action?

Evidence to collect: Versioned policies and reusable checks, with exceptions and authorized scope.

Inspect behavior

Can you trace a decision through the inputs, tool actions and saved effects?

Evidence to collect: Execution identity, source lineage and an action journal, including forbidden attempts.

Handle exceptions

Who decides whether to contain, correct, approve or escalate a failure?

Evidence to collect: A named owner, recorded rationale, effective version and follow-up check.

Maintain the evidence

What causes review when tools, permissions, policy or agent behavior change?

Evidence to collect: Freshness records, drift signals and a defined review trigger; record unknown coverage.

Turn the gaps into a bounded next step

Start with the most consequential unresolved job. Agree what would establish a better result, test the correction, and retain the source and outcome limits.

Use the reliability workflow worksheet · Explore Tensile · Discuss a concrete workflow