Core Concepts

Optra Prism is built on a few core ideas.

The problem

AI coding agents are powerful but opaque. Users face:

• No prompt feedback — you don’t know if your prompts are efficient or wasteful
• Invisible throttling — rate limits silently slow you down with no visibility
• No cost visibility — token spend accumulates with no breakdown by session, model, or pattern
• No coaching — you repeat the same mistakes because nothing tells you what to improve
• No enforcement — Prism measures and coaches; it doesn’t block, gate, or cap anything

Prism solves these by instrumenting the AI coding workflow and surfacing insights.

The north star

A good prompt — and a good sub-session — reaches the goal using the minimum necessary tokens, turns, human think time, and response latency, given the current session context.

Those four fundamental values drive every score Prism computes.

The Prompt Score

Prism collapses “how well is this user directing the AI?” into one headline number — the Prompt Score:

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

Every prompt is graded by seven rubric checks applied across eight intent classes, rolled up to a letter grade; averaging those grades gives the Prompt Score. It’s the number at the top of your dashboard.

Three supporting pillars break the headline down, and a Completion rate tracks outcomes alongside it:

Metric	Question	Unit
Prompt Score	How well do you direct the AI, prompt by prompt?	Average prompt grade, 0–100
Speed	How much finished work do you crush per window?	Crushed sub-sessions/window
Skill	Are your good prompting habits consistent?	Skill Index, 0–100
Token Usage	How many tokens does each crushed sub-session cost?	Tokens/crushed sub-session (lower is better)
Completion rate	What share of scored sub-sessions crushed their goal?	crushed ÷ scored sub-sessions

A sub-session crushes when it clears the substance floor, reaches its goal, and is not redone later — confirmed by an LLM outcome judge against per-intent criteria. Completion rate is a supporting outcome metric, not the headline.

See The scoring model for how the Prompt Score and its supporting metrics are computed.

Three layers, one pipeline

Prism’s data model is three measurement layers plus a parallel intelligence pipeline:

• Layer 0 — Telemetry. OpenTelemetry from Claude Code.
• Layer 1 — Facts. Four things established per sub-session: whether the goal was reached, whether it had to be redone, how many tokens it took, and how well the prompts were written.
• Layer 2 — Scores. The headline Prompt Score (average prompt grade, 0–100) plus the supporting Completion rate (crushed ÷ scored sub-sessions).
• Prompt rubric. Grades every prompt against 7 quality checks, applying only the ones that fit what you were trying to do.

Intelligence loop

Prism creates a feedback loop between coding and coaching:

Code with AI  →  Capture telemetry  →  Score & analyze  →  Surface insights  →  Improve  →  repeat

• Real-time: the dashboard realtime view scores each prompt as you send it and surfaces a per-sub-session Task Advisor
• Session-level: the backend grades each prompt and scores each sub-session
• Trend-level: the dashboard shows improvement over days and weeks
• Rightsizing: an estimate, from your usage patterns, of which subscription tier you need