Prompt Score

Prompt Score is your personal headline score. It leads with one direct question about your prompting:

How good, on average, were your prompts?

Your Prompt Score is the Average Prompt Grade (APG) rendered on a 0–100 scale with a letter grade. Every prompt is graded against seven rubric checks across eight intent classes, mapped to a letter grade, and averaged into the headline number.

Prompt Score = Average Prompt Grade (0–100), shown with a letter grade

The hub page /score-v3 leads with that number. Three cards sit beside it:

• Trivial sub-sessions — how many the substance floor held out of scoring.
• Merges — how many the anti-fragmentation guard collapsed into one, so a stop-and-resume isn’t counted twice.
• Crush Weight — the median tokens it takes to finish one crushed sub-session.

Below the cards the hub draws your Prompt Score over time, a comparison against your teammates, and the table of sub-sessions behind the score.

Sub-sessions and crushing

The scored unit underneath the Prompt Score is the sub-session — one goal arc within a Claude Code session, closed by /clear or a topic shift. Sub-sessions are what feed the supporting Completion rate.

A sub-session “crushes” when all three of these hold:

1. Substance floor passed — the sub-session did real work (≥3 turns, or ≥10 net LOC, or ≥1 mutating tool call)
2. Goal complete — verified by an LLM outcome judge against per-intent criteria
3. Not rework — no later sub-session reverts or rewrites the same code

The Completion rate is crushed ÷ scored over sub-sessions that passed the substance floor. Sub-sessions that fail the substance floor or never establish a goal are excluded from both numerator and denominator — trivial chat neither drags the rate down nor pads it.

The Completion rate has no card of its own on the hub. You meet it on the Sessions table (which marks each row as counted or excluded), in your Report, and across the Team Console.

Session labels

Every sub-session also gets two human-readable labels:

• Title — up to 60 characters (e.g. “Fix off-by-one in JWT parser”).
• Summary — a 1–2 sentence recap, up to 240 characters.

Labels follow a three-step lifecycle:

1. Heuristic — when the sub-session opens, a title is taken from a truncation of the first user prompt.
2. Rubric — when the sub-session closes, the rubric LLM call rewrites both fields with a cleaner title and summary.
3. User-edited — you can rename a sub-session in the dashboard at any time. A user edit locks it against future overwrites.

What’s in this section

This section explains how the scoring works:

• Algorithm Overview — sub-session boundaries, the substance floor, the crushed-sub-session rule
• Prompt Grade — per-prompt rubric grading and the Average Prompt Grade that drives the headline
• Trivia — sub-sessions filtered out by the substance floor
• Integrity — anti-gaming flags raised against your sub-sessions
• Advisor & Session Summaries — LLM-generated session summaries and prompt-rewrite advice

The screens themselves sit in the dashboard’s Analytics group: Prompt (the hub), Speed, Token Usage, and Skill. What Prompt Grade, Trivia and Integrity describe shows up on the hub and on the Sessions screen. See the Prompt Score hub for the map.