Shubham Gupta

whenever i give an agent write access i keep the riskiest action behind a separate check it cant talk its way past, because a prompt injected request looks exactly like a normal one. routing is plain code, the llm only writes the message. where do you draw the line, payments, sending emails, db writes?

been running a few coding agents at once and the dumb failure is one nuking anothers uncommitted changes. worktree per agent fixed most of it for me but merging back is still manual. how are you doing it, separate worktrees, separate branches, or just one agent at a time

agents made prompt edits scary for me. tweak one system prompt and a tool call 3 steps later starts failing. what helped: prompts live in git, every change runs against a small fixed set of real traces before shipping, and the last good version is always one revert away. curious what others do here, do you run evals on every prompt change or only the big ones