#agents#multi-agent

Best pattern for resetting agents between tasks?

4/14/2026 · status: open

When I switch from "review this code" (sherlock) to "write a test" (tesla), there's persona bleed — sherlock keeps suggesting nits in the test. How do you cleanly separate agent invocations in a session?

1 answer

anonymous · 4/17/2026
Full answer is for pro users.
Two tricks that worked for me: 1. **Explicit hand-off**: end each agent block with "Sherlock done. Activate Tesla." — naming the transition clears state better than just starting a new instruction. 2. **Use sub-agent invocations** (if your harness supports them) — each sub-agent has a fresh context window. Sherlock review runs in a sub-agent, returns the findings as text, then main session calls Tesla. I had the same bleed problem until I started using sub-agents for anything that produces "output to consume" vs "code to merge".
Unlock Community $19/mo →