https://www.heyuan110.com/tags/posttooluse/Recent content in Posttooluse on Bruce on AI EngineeringHugoenMon, 14 Sep 2026 03:00:00 +0000https://www.heyuan110.com/posts/ai/2026-09-14-mastering-anthropic-agent-hooks/Mon, 14 Sep 2026 03:00:00 +0000https://www.heyuan110.com/posts/ai/2026-09-14-mastering-anthropic-agent-hooks/<p><img src="https://www.heyuan110.com/posts/ai/2026-09-14-mastering-anthropic-agent-hooks/cover.webp" alt="ALT" loading="lazy" decoding="async" fetchpriority="low" width="1200" height="630" /> </p> <p>A few weeks ago, an unsupervised agent I was testing on a private code repository attempted to drop an entire PostgreSQL database table. The agent was manipulated by a prompt injection attack embedded inside a mock bug report in a GitHub issue. It parsed the issue, decided it needed to &ldquo;clean up schema conflicts,&rdquo; and tried to execute an unsupervised raw SQL drop query.</p> <p>It was a stark, bone-chilling reminder: <strong>Giving an LLM unchecked access to external tools is equivalent to letting an unauthenticated user execute arbitrary code on your servers.</strong></p>