Sprint Eval

Panel-based sprint plan rating. Three independent expert agents read the SAME plan and score the SAME criteria, so scores are comparable and agreement is signal: anything all three flag independently is almost certainly real.

Input

A sprint plan: a file path or pasted text. If it's a file, pass the path to the agents and let each read it themselves (independent reads, no shared summary — don't pre-digest the plan for them, that would correlate their errors).

Fixed criteria (same for every rater)

1. Dependency & sequencing health — critical path soundness, blocker surfacing/ordering, cross-person handoffs.
2. Capacity realism & load balance — per-person allocation vs the sprint horizon, onboarding load, P0 distribution.
3. Goal alignment & scope discipline — must-ship items map to stated targets; cancel/backlog/carry-over hygiene; clarity.

The panel

Spawn all 3 in ONE message (parallel Agent calls, general-purpose). Each gets a distinct lens so they catch different failure modes: