Commands Reference

/prism:setup

Configure your API key and enable telemetry.

/prism:setup gck_YOUR_API_KEY

What it does:

Validates the key format (must start with gck_)
Calls the config endpoint to resolve service URLs (ingest, gateway, dashboard)
Caches the resolved config locally (24h TTL)
Creates ~/.prism/ directory (mode 700) and writes config (mode 600)
Syncs OTEL telemetry env vars to ~/.claude/settings.json

The API key is also stored in Claude Code’s native userConfig system (system keychain for sensitive values).

After setup: restart Claude Code for OTEL telemetry to take effect.

/prism:status

Show connection health, active features, gateway state, and status line toggle.

/prism:status

Displays:

API key prefix (e.g., gck_abc12...)
Gateway routing mode (enabled/disabled with URL)
Status line state (on/off)
Active environment variables (PRISM_INGEST_URL, ANTHROPIC_BASE_URL, OTEL endpoints)
Active features: OTEL telemetry, PRISM gate, prompt capture, gateway routing
Current session turn count

Toggle gateway routing:

/prism:status toggle

Flips enableGateway in your config and prompts you to restart.

Toggle status line:

/prism:status toggle status line

Flips showStatusLine in your config. Takes effect on the next response.

/prism:cost

Show current session cost and token usage.

/prism:cost

Displays:

Number of turns in current session
Session duration (time since start)
Estimated session cost (cache-aware pricing when OTEL data is available)
Reminder to use /compact if turns exceed 10
Tip to use /model sonnet for simple tasks if using Opus
Link to dashboard for detailed cost breakdown

Cost is computed using model-specific pricing with cache-aware rates. The stop hook queries OTEL telemetry for cache_read_tokens and cache_creation_tokens to calculate accurate costs. When no cache data is available, full input pricing is used as a conservative estimate.

/prism:score

View your PRISM profile with coaching tips.

/prism:score

Queries the engine’s /v1/intelligence/prism endpoint for the last 7 days of scores.

Displays four sections:

Weakest dimension — sub-scores with 3 coaching tips and examples
Bar chart — all 6 dimensions visualized (filled/empty blocks), weakest and strongest marked
Optimization tips — 2–3 actionable improvements based on your profile
Dimension reference — all 6 dimensions with weights

Dimensions and weights:

Dimension	Weight	Metrics
PQ (Prompt Quality)	25%	Specificity + Decomposition
IE (Iteration Efficiency)	20%	Convergence + Recovery
VD (Verification Discipline)	20%	Review + Validation
CQ (Code Quality)	15%	Judgment + Safety
TU (Tool Use)	10%	Selection + Context
AF (Advanced Features)	10%	Delegation + Configuration

Proficiency scale: 9.0+ Elite, 7.0–8.9 Expert, 5.0–6.9 Proficient, 3.0–4.9 Practitioner, <3.0 Novice

/prism:report

Comprehensive 30-day review of trends, habits, and waste.

/prism:report

Queries PRISM scores (last 30 days, up to 1000), waste patterns, and pre-generated insights reports.

Displays six sections:

Overview — total turns, sessions, date range, overall PRISM average, proficiency tier
Trends — per-dimension trajectory (first week vs. last week), labeled improving/slow/stable/declining
Strengths & Weaknesses — dimensions above 7.0 and below 5.0
Habits — correction rate (target <10%), top 3 recurring friction patterns
Worst prompts — 5 lowest-scoring turns with identified patterns
Waste — top 3 waste patterns by severity with estimated savings

Ends with a single focus area recommendation and a link to the dashboard.

/prism:advisor

Prompt optimization advisor with detailed analysis and multilingual support.

/prism:advisor "your prompt text here"

Scores the prompt across all 6 PRISM dimensions (conceptual evaluation, works in any language).

Displays:

PRISM score (X.X/10) with all dimension sub-scores
Bar chart of all 6 dimensions
Weakest and strongest dimensions
Original prompt
Suggested rewrite using real file paths from the project
Why the rewrite is better

Rewrite rules applied:

Add real file paths from the project (not placeholders)
Add function/class names for specificity
Remove bundling phrases (“and also”, “while you’re at it”)
Add scoping language (“only”, “just”)
Add expected behavior / acceptance criteria
Split multi-task prompts to first task only

Anti-patterns flagged:

Retry storms — suggests adding constraints instead of repeating
Context dumping — suggests asking which section matters
Multi-task bundling — suggests splitting into sequential prompts
Blind acceptance — encourages verification after changes
No planning — suggests /plan for complex tasks
Missing CLAUDE.md — suggests creating a project context file

Session context: The advisor reads ~/.prism/advisor-context.json (updated by the submit hook) for turn count, token growth, model usage, and response times to tailor its advice.

/prism:help

Show all available commands grouped by category.

/prism:help

Output:

Prism Plugin Commands

Setup & Config
  /prism:setup [gck_KEY]   Register API key, enable telemetry + gateway
  /prism:status            Connection health, gateway toggle, session info
  /prism:uninstall         Remove plugin, clear all settings

During a Session
  /prism:cost              Session cost, token usage, compact tip
  /prism:score             Weakest dimension, coaching tips, optimization advice
  /prism:advisor [prompt]  Optimize a prompt — PQ score, rewrite, and tips

After a Session
  /prism:report            Full review — trends, habits, waste, worst prompts

Automatic (hooks — no command needed)
  Prompt advisor   Scores prompts in realtime, blocks low PQ with rewrite advice
  Response timer   Shows elapsed time and token count after each response
  Context nudge    Smart /compact and /clear advice based on context growth
  Waste alerts     Model overkill warnings every 10 turns

/prism:uninstall

Remove the plugin, configuration, and OTEL settings.

/prism:uninstall

Steps:

Asks for confirmation
Removes OTEL environment variables from ~/.claude/settings.json
Deletes ~/.prism/ directory (config and cache)
Reminds you to restart Claude Code

After uninstalling, OTEL telemetry export stops and all plugin hooks deactivate.