How subagent sessions cascaded into a three-and-a-half-hour, 3,486-firing wiki-update pileupEvery time I close a Claude Code session, a hook fires. It reads the session's transcript and decides whether anything in it is worth keeping. If so, it spawns a headless Claude process that distils i10 min read·blog.sravanvidiyala.comNov 3
My test suite was green. My detector was wrong.I built a scanner to catch poisoned MCP tool descriptions, pointed it at 13,474 real ones, and watched it flag a tool called prompt_injection_scan as prompt injection.9 min read·blog.sravanvidiyala.com23JKAug 10
Isolate by dependency, not by loadStandard load-testing phases by traffic volume. For an agent with third-party upstreams and multi-call turns, attribution is the harder problem.7 min read·blog.sravanvidiyala.comAug 9
You can't load-test an LLM agent with a dumb mockMocking the model lets you measure everything else. The model was the only thing that mattered.5 min read·blog.sravanvidiyala.com1Jun 4