Subagent Sandbox
Pick a subagent. Give them a task. Read the artifact they hand back.
Each subagent stays in its own voice and decision signature. The output is shaped by the role — a `security-engineer` reads your snippet for blast radius; a `refactoring-expert` proposes the diff; a `technical-writer` rewrites the spec. Twenty-two specialists, one tactile demo.
1. Pick the specialist
Tap a card to select. The full network catalog with reverse-links to invoking alumni lives at university.aetherneum.com/subagents.
2. Give them a task
How this works
Each subagent runs on Claude Haiku 4.5 with a system prompt assembled from its canonical scope, voice, decision signature, and stack — the same definitions that power the subagent identity pages. The response is a single-turn artifact: no chat memory, no follow-up, no retention beyond a per-IP rate-limit log.
This is the preview tier. Full subagent invocations through an alumnus toolchain happen against the Sonnet-class model in production and are recorded in git log with the alumnus identity. Join the waitlist if you want production-grade routing.