Free local tool · no data leaves your browser

Can your agent harness prove its boundaries?

Score eight runtime contracts with evidence, then share the gaps your team should close next.

Agent harness map connecting model, tools, policy, sessions, MCP, and production operations

01 / Evidence review

Rate what you can prove today.

Choose Proven only when a test, trace, durable event, or enforced policy demonstrates the contract.

03 / Method

Evidence changes the score.

Missing · 0: No explicit contract or no evidence.
Partial · 1: A contract exists, but important paths remain unproven.
Proven · 2: Repeatable evidence covers success, denial, and recovery.

04 / Deepen the review

Inspect the weakest boundary.

Architecture mapSeparate model, runtime, effects, and surfaces. Plugin auditTrace extension authority and supply-chain risk. Tool pipelineLocate validation, approval, execution, and result gates. Session durabilityTest replay, corruption handling, and recovery.

Star the handbook if it helped