What is starlight-eval?

An evaluation command for checking an intelligence system across separate areas such as its model, memory, search, testing setup, underlying software, datasets, and overall results. It produces findings for each area and a combined scorecard.

Use it to run all evaluation lanes or just one, reuse existing tests and benchmarks, review dataset origins, and record the results.

It gives one repeatable way to see where the system works and where it needs attention, including changes since an earlier run.

How frankxai/Starlight-Intelligence-System configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to frankxai/Starlight-Intelligence-System. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/frankxai/Starlight-Intelligence-System/main/.claude/commands/starlight-eval.md

Clone the repo

git clone --depth 1 https://github.com/frankxai/Starlight-Intelligence-System

Made for: Claude Code.

Or install starlight-intelligence-system, the plugin that ships this one along with the rest of its 5 skills, 121 commands, 7 agents.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

README.md

[![agentmods](https://agentmods.dev/badge/commands/frankxai/starlight-intelligence-system/starlight-eval.svg)](https://agentmods.dev/commands/frankxai/starlight-intelligence-system/starlight-eval)

Your own site

<a href="https://agentmods.dev/commands/frankxai/starlight-intelligence-system/starlight-eval"><img src="https://agentmods.dev/badge/commands/frankxai/starlight-intelligence-system/starlight-eval.svg" alt="Measured on agentmods" height="20"></a>

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 809 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5.1	$0.00000	$0.00809
Opus 5	$0.00000	$0.00404
Sonnet 5	$0.00000	$0.00162
Haiku 4.5	$0.00000	$0.00081

Measured 3d ago against content hash 0bfc88f65609, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

starlight-eval scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/commands/starlight-eval.md · 61 lines

How it starts

The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/starlight-eval

Run a Starlight Proving Ground pass — evaluate the whole Starlight Intelligence System across seven lanes, render per-lane and system verdicts with the Luminor kernel mindset, and write a scorecard receipt.

Spec: tools/proving-ground/SPEC.md · Lanes: tools/proving-ground/lanes.json Evaluator disposition: agents/starlight-evaluator.md

Usage

/starlight-eval              # full pass — all 7 lanes
/starlight-eval <lane>       # single lane: model | memory | retrieval | harness | substrate | datasets | system
/starlight-eval --since <tag-or-runId>   # delta against a prior scorecard

What it does

Load lanes.json. For each requested lane, run or read its entrypoint (compose existing infra — do NOT re-implement measurement):
- model → run an arena round (tools/arena/) or read the latest receipt.
- memory → run bencher.py if corpus ≥ 50, else read last memory/benchmarks/.
- retrieval → run test/retrieval-eval.test.ts.
- harness → run tools/run-v01-evals.mjs.
- substrate → run npm run test:substrate.
- datasets → audit dataset provenance (verdict, not metric).
- system → synthesize lanes 1-6.
Evaluate each lane with the Luminor kernel mindset (Precision / Wisdom / Transcendence). Per lane: a board verdict (PROCEED / REVISE / STOP), the metrics with sourceLane + baseline + delta, caveats, and the named weakness the passing numbers hide (Wisdom layer — required, never omit).
Render the system verdict via the Starlight Board pressure-test (/starlight-board) over the composed lane results.
Write the scorecard to tools/proving-ground/scorecards/<runId>.json with ranAt, nextRunDue, cadence, and the antiGoodhart warning (R1+R2 bindings).
Attest — embed "Built on SIP" in the scorecard. If publishing, the artifact is Board-gated before it reaches /research.

Rules

Compose, never duplicate. Every metric traces to an existing lane entrypoint.
Mechanically verified beats judged. Prefer asserts/ground-truth; when a judge is unavoidable, use a blind non-contestant (cross-family judge preferred) and say so.
Name the weakness. A lane verdict without a named weakness is incomplete — the Wisdom layer is the point, not the score.
Do not optimize to the score. The scorecard describes the system; it is not a target. State this in every published artifact (anti-Goodhart, R2).
Staleness is visible. nextRunDue past today → the public surface shows STALE.
This command is substrate-tier: a /starlight-eval pass is part of board-before-tag for substrate releases.

Read the full file on GitHub · 61 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

3d ago First seen · 61 lines · 0 tokens per session scan A 0bfc88f65609

Subscribe to this mod's changes

starlight-eval is a command published in the GitHub repository frankxai/Starlight-Intelligence-System (8 stars, last pushed today), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 809 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Analytics and metrics Claude Code Memory and context Multi-Agent Orchestration TDD Testing ai-agents intelligence

Starlight-Intelligence-System: Command for Claude Code

Borrowing it

Wrote this? Show the measurements

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

/starlight-eval

Usage

What it does

Rules

What this file has done since we first saw it

Other commands, from other repositories

prompt-evolve

hatch3r-bug-pipeline

bugfix

fix

manage-skills

lint-harness