Aftermath logo

Aftermath

Outcome-label eval kit for agent guardrails. by Haku

Artificial Intelligence Developer Tools Open Source

Guardrails classify tool names; incidents classify outcomes. Aftermath ships the labeled data Oct 2026 news demanded: 30 agent traces — the HTTP-GET-that-writes, the decrypt-after-filter exfil, the citation-tool proxy, the test-env escape — labeled safe / data-exfil / unauthorized-write / scope-escape / resource-abuse. Outcome classifier: 30/30 on the corpus; tool-name allowlist: 8/30. Includes the stdlib CLI, runbook, and one-page evidence report for the NY RAISE Act 72h clock (Jan 1, 2027).

投票数: 0
← 投稿一覧に戻る