Plugin Overview

The Prism plugin runs inside Claude Code and provides four automatic hooks, eight slash commands, and an always-active prompt advisor skill.

How it works

The plugin activates on every Claude Code session:

1. Session start — validates your API key, resolves service URLs from the config endpoint, syncs OTEL settings, resets session state
2. Before each prompt — captures prompt text, scores PQ quietly (for the status line), shows context management nudges
3. After each response — shows a status line with response time, tokens, cost, PQ score, and turn count
4. After /compact — resets all session counters for fresh tracking
5. Always-active advisor — the prism-advisor skill evaluates every prompt and shows one-line coaching when improvement is possible

Automatic features (hooks + skill)

These run without any commands — they’re always active:

Hook	Trigger	What it does
Session start	Session begins	Validates key, resolves URLs, syncs OTEL, resets counters
Prompt handler	Before prompt submit	Captures prompt, scores PQ quietly, context nudges
Status line	After response	Shows [Prism] 3.2s · 12K in / 2K out · $0.23 ($1.45) · PQ 7.5 · turn 5
Post-compact	After /compact	Resets turn count, cost, and context tracking
Prism advisor (skill)	Every prompt	One-line coaching for low-specificity, bundled tasks, retry storms

Slash commands

Run these anytime during a session:

Command	Purpose
/prism:setup [key]	Configure API key and gateway
/prism:status	Connection health, gateway toggle, status line toggle
/prism:cost	Session token usage and cost
/prism:score	PRISM profile with coaching tips
/prism:report	30-day comprehensive review
/prism:advisor [prompt]	Detailed prompt optimization analysis
/prism:help	List all commands
/prism:uninstall	Remove plugin and settings

See Commands for detailed reference.

Session state

The plugin tracks state across your session in ${CLAUDE_PLUGIN_DATA}/session-state.json:

• Turn count and session duration
• Cumulative input/output tokens and estimated cost
• Context growth (first vs. current input tokens)
• Model usage counts and cache rate from OTEL
• Response times (last 50)
• Last PQ score (from submit handler, displayed on status line)

This resets on session start and after /compact.