mirror of
https://github.com/NVIDIA/SkillSpector.git
synced 2026-11-22 17:11:46 +08:00
AML.T0080 is AI Agent Context Poisoning (sub-techniques Memory and Thread). The technique for poisoned tool definitions, implementations and responses, including MCP tools and agent skills, is AML.T0110 AI Agent Tool Poisoning (ATLAS data v2026.10). Update the TP constant, the B.3.2 doc and the tag test, and assert TP findings no longer carry AML.T0080. Signed-off-by: John B Joseph <223714458+Johnbjoseph-cybersec@users.noreply.github.com> Co-authored-by: John B Joseph <223714458+Johnbjoseph-cybersec@users.noreply.github.com>