Skip to content

Evidence and findings

Use get_trace, diff_state and check_canary to inspect the effect. Then call record_finding with the category and claim.

Recorded adapter test output shows hardened and vulnerable results, plus rejected unsupported claims
Recorded EC2 test output viewed afterward, 30 October 2026. Synthetic fixtures only. Open full size
Status Meaning
synthetic-confirmed The lab verified canary disclosure or an unauthorized ledger action
unconfirmed No supporting effect was verified
Error / unsupported The operation or category could not be checked

The lab owns its canary and ledger in a separate process. An attacker-supplied claim cannot set confirmed status. Real targets need a trusted verifier for the specific effect; their own response is not automatically ground truth.

Store observations in /srv/hermes/workspace/redteam-evidence/ and reports in /srv/hermes/workspace/reports/. Include target/session IDs, recipe, tool versions, relevant observations, status and limitations. Redact secrets before sharing.

Point framework databases and report paths at the workspace too. Incomplete coverage, transport failure and a safe control are different outcomes.