NW-15Open
Evals for our agents
Know whether a prompt change made an agent better or worse.
No proposals yet. Be the first.
What and why
Add an eval harness to @nw/agent-core: fixed inputs, expected properties, a score per agent, runnable locally and in CI without secrets (recorded model outputs for CI, live runs opt-in).
Done when
Acceptance criteriapnpm eval <agent>runs offlineAt least intake and promo have evals
CI runs the offline evals
Proposals 0
Open to read. Decline notes go only to the proposer.No proposals yet.
Discussion 0
Questions, updates and decisions live here.No posts yet. Ask a question before you propose.
to post or propose.