Prompt Grade

Prompt Grade scores each prompt you send against a seven-point rubric. The judge runs after the prompt is captured and produces a single letter grade from A+ down to F.

Individual grades appear in the Grade column of the Sessions table — open a row to see which rubric checks the prompt passed. Their average is the score hub’s headline Prompt Score.

How a prompt is graded

Each prompt is checked against seven habits, each rooted in Anthropic’s Claude Code prompting guide. Not every check applies to every prompt — the judge marks which checks are applicable based on intent (fix, plan, explore, etc.).

The seven checks roll up into four pillars (Clarity · Context · Verification · Workflow) — see the Skill page for the full list.

APG — Average Prompt Grade

Each prompt’s rubric score is a 0–100 percentage — the share of applicable quality checks it passed — which then maps to a letter grade (see Bands below), and each letter is worth grade points (A+ = 11 down to F = 0). For each sub-session, Prism takes the grade points of its best prompt; APG is the average of those per-sub-session bests across the sub-sessions that cleared the substance floor, rescaled to 0–100.

APG = the average of each sub-session's best prompt grade (in points), over sub-sessions that cleared the substance floor, rescaled to 0–100

The headline number is APG; the badge next to it is the closest letter grade. This is the same number the /score-v3 hub surfaces as your Prompt Score.

Bands

Poor under 36 · Weak 36–64 · Fair 64–82 · Good above. These are the same four names the /score-v3 hub puts on the same number, so a given APG never reads under two labels.

Each prompt’s own rubric percentage maps to a letter separately: A+ ≥ 90, A ≥ 80, A- ≥ 72, B+ ≥ 62, B ≥ 50 (the baseline), B- ≥ 42, C+ ≥ 34, C ≥ 24, D ≥ 12, and F below 12.