# Log

## [2026-06-10] ingest | A2A, OpenClaw, skill, CUA security paper set

- Created `llm-wiki-a2a-agent-security` under `docs/presentation-0611/`.
- Collected the user-provided starting literature map.
- Collected 11 paper sources and linked their public archive pages in `sources.md`.
- Created source pages, concept pages, index, schema, and presentation-oriented synthesis.

Source grouping:

- A2A protocol security: A2ASecBench, AI-agent protocol threat modeling.
- Local personal agent execution risk: OpenClaw security analysis, ClawWorm, FASA/ClawGuard.
- Agent skill ecosystem risk: Skill-Inject, PoisonedSkills/DDIPE, malicious skills in the wild.
- Computer-use agent risk and defenses: CaMeLs/NOVA, AgentSentinel, RedTeamCUA.