The audit is the trust product.
A multi-agent run is only useful if you can trust the result. Voss makes the audit trail the center of the product: a replayable, deterministic record of every action, decision, and gate outcome — good enough to review a PR from.
What it answers
Understand a run without the logs.
What actually changed
Diff summary, agent actions, and tests per card — understandable without reading raw logs.
Claims vs. evidence
The audit separates what the EM said from what was independently verified. Unsupported claims are detectable.
Why it passed — or didn't
Reviewer-A verification and Reviewer-B verdict are recorded separately, linked to files, tests, and eval output.
What's still risky
Every run ends with a residual-risk section and the blocked, killed, and rescoped lineage.
The report
Fifteen sections, one trail.
Every audit is deterministic from persisted run data and exports to Markdown and JSON. In the ADE it renders as a navigable session tree.
Built for review and replay.
Use the audit to review a multi-agent run like a PR, to detect unsupported claims, or to replay exactly what happened. Budget allocation and consumption are shown per node, alongside scope violations and denied attempts.
- $ voss review <run_id>Reviewer-A + Reviewer-B per card
- $ voss board <root_id>Card state from persisted nodes
- $ voss session tree <root_id>Per-agent budget, scope, status
- $ voss team run "<goal>"Run that produces the trail