Free core · northern forge

Agent Eval Rubric Desk

Score goal, tools, evidence, and next fix → one paste-ready pass/fail rubric card after an agent tool session.

Use freeFull core runs in your browser — no account required.

ProFree core + $3.99 pro. One-time unlock on this device.

Local-firstYour data stays in the browser unless you export it.

Session goal Agent / model / stack Tools actually used (comma or one per line) Evidence / proof What failed / open Next fix

Goal met

Tool use

Evidence

Safety / no thrash

Scope discipline

Next-fix quality

Verified with a real command or live URL (not just model claim) No secrets pasted into public surfaces Needed human rollback / emergency fix Hallucinated tools, URLs, or APIs

Pass threshold (of 12)

Card format

Pro: multi-session scoreboards, team rubrics, and export to Linear/GitHub issues.