judge

A manual review command for scoring how well a coding skill or agent performed. It uses a seven-part quality review and can compare the result with an earlier score.

In plain words
What is it for?
Use it after running a skill or agent to evaluate execution quality, choose a specific review rubric, or compare the result with a previous evaluation.
Why use it?
It gives a structured way to find weaknesses in an agent run instead of relying only on a general impression. It can also account for different agent systems and review rubrics.

Command

Part of the proofloop plugin — 1 skill, 6 commands, 1 agent, 3 hooks shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/sattyamjjain/proofloop/judge
Clone the repo
git clone --depth 1 https://github.com/sattyamjjain/proofloop

Or install proofloop, the plugin that ships this one along with the rest of its 1 skill, 6 commands, 1 agent, 3 hooks.

Per session 11 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,757 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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.

ModelPer sessionOnce invoked
Fable 5 $0.00011 $0.01757
Opus 5 $0.00005 $0.00879
Sonnet 5 $0.00002 $0.00351
Haiku 4.5 $0.00001 $0.00176

Measured 3d ago against content hash fa2d2746ba6e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

judge 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.

commands/judge.md · 121 lines

How it starts

The opening of the file, as written. The whole thing — 121 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/judge — Manual Skill Quality Evaluation

You are Proofloop, the universal quality evaluator for Claude Code skills and agents.

Your Task

When the user invokes /judge, evaluate the most recent skill or agent execution using the 7-dimension scoring system.

Arguments

  • skill-name (optional): The name of the skill to judge. If omitted, detect the last skill that ran from the conversation context.
  • --rubric RUBRIC (optional): Use a specific rubric file (e.g., security, code-review). If omitted, auto-detect the best rubric.
  • --verbose (optional): Show detailed per-dimension justifications.
  • --adapter NAME (optional): Transcript adapter for non-native ecosystems. One of claude-code, cowork, openai-compatible, codex, cursor, continue.
  • --model ID (optional): Model ID override (e.g., claude-opus-4-7). When omitted, the model is auto-detected from the transcript and the tokenizer_baselines config scales efficiency length thresholds accordingly.
  • --against REF (optional): Compare the current scorecard against a previous run of the same skill (HEAD~1 = penultimate scorecard, numeric index = absolute). Delegates to skills/judge/scripts/against.py and exits non-zero on composite regression.
  • --watch (optional): Run the live re-scoring daemon in skills/judge/scripts/watch.py. Polls the scores directory every 2 s, prints a one-line diff header per change (improved X, regressed Y, unchanged Z since last run), and re-emits Proofloop Studio to the --output HTML path. Pairs with /scorecard or /benchmark during iterative skill development.
  • --llm-second-opinion (optional): Force the opt-in LLM second-opinion analyzer on for this one run even when llm_second_opinion.enabled = false in judge-config.json. Requires ANTHROPIC_API_KEY. See skills/judge/analyzers/llm_judge.py.
  • --explain SCORECARD (optional): Render an existing scorecard JSON in either Markdown (PR-friendly) or JSON (CI-friendly). Delegates to skills/judge/scripts/explain.py. Pair with --format md|json (default md) and optional --out PATH. Example: /judge --explain skills/judge/scores/code-review_2026-04-25.json --format md. See SKILL-judge-explain.md for the output schema.
  • See also /compare for explicit two-file delta with Auto Memory regression narrative — complement to /judge --against HEAD~1.

Read the full file on GitHub · 121 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.

  1. 3d ago First seen · 121 lines · 11 tokens per session scan A fa2d2746ba6e

Subscribe to this mod's changes

judge is a command published in the GitHub repository sattyamjjain/proofloop (5 stars, last pushed 2mo ago), licensed MIT. It adds 11 tokens to every session and 1,757 once invoked, about $0.0001 per session on Opus 5. 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 2026-08-31.

Related

Other commands, from other repositories

harness-gate

Gate the agent setup on corpus-validated rules (gating tier). Fast, no LLM, exits nonzero on any finding. Suitable for CI and pre-commit.

redhat-community-ai-tools/harness-eval · 0 tokens

test

Run Postman collection tests, analyze results, diagnose failures, and suggest fixes.

Postman-Devrel/postman-claude-code-plugin · 16 tokens

routine

Execute one maintenance routine defined in scv/routines/ .md (task + guardrails + exit-criteria contract), or list defined routines. SCV never schedules — pair with host features like /loop or cron yourself. Use whenever the user asks to run a recurring maintenance task, or asks what routines exist — not only when…

wookiya1364/scv-claude-code · 75 tokens

workspace

Set up or change this repo's place in a multi-repo SCV workspace: join an umbrella as a child, create an umbrella (root), check status, or detach. Friendly — no long flags to remember. Use whenever the user asks how this repo relates to sibling repos, or wants to join, create, or leave a multi-repo workspace — not…

wookiya1364/scv-claude-code · 83 tokens

incident

Live incident response — intake a Sentry issue, stack trace, or symptom; gate mitigate-first vs root-cause-first; keep a UTC timeline. Use when production is broken RIGHT NOW; for the after-action write-up alone, use /orc:postmortem.

HigorAlves/orc · 54 tokens

t800-onboard

Для чата с новичками. Показывает, что настроено в Cursor (global + local), и что умеет отдел T-800.

Khar-AG/t-800-agent · 0 tokens