CODING AGENT WORKFLOW · PRACTICAL GUIDE

Completion Claim Evidence Check

Compare completion claims with caller-supplied text overlap and optional reported test exit codes, then call out missing or conflicting evidence.

THE PROBLEM

An agent can say that a change is finished even when it did not run the test command or inspect the resulting diff. A small referee can ask for the trace behind that claim.

HOW THIS TOOL WORKS

completion_verify reports overlapping claim terms with the text-overlap-only or unverified status, then reports optional caller-supplied test-result exit codes separately. It labels a zero exit as reported-success, not as an independently verified test run.

TRY IT

Provide the completion claim and a redacted test-result excerpt with kind=test-result and its reported exitCode.

LIMITS

Evidence fields and exit codes are caller-supplied and unauthenticated. The tool does not execute commands, inspect the actual diff, or prove behavior. Never mark a task complete solely because this tool returned an overlap or reported-success status.