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.
npx agentmods add skills/sattyamjjain/proofloop/judgenpx skills add sattyamjjain/proofloop --skill judgegit clone --depth 1 https://github.com/sattyamjjain/proofloopWrote 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.
[](https://agentmods.dev/skills/sattyamjjain/proofloop/judge)<a href="https://agentmods.dev/skills/sattyamjjain/proofloop/judge"><img src="https://agentmods.dev/badge/skills/sattyamjjain/proofloop/judge.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00022 | $0.04292 |
| Opus 5 | $0.00011 | $0.02146 |
| Sonnet 5 | $0.00004 | $0.00858 |
| Haiku 4.5 | $0.00002 | $0.00429 |
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 4d 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.
How it starts
The opening of the file, as written. The whole thing — 382 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Proofloop — Universal Quality Evaluator
Overview
Proofloop is a universal quality evaluator for Claude Code skills and agents. It measures execution quality across 7 weighted dimensions, producing an evidence-based scorecard with letter grades, justifications, and actionable recommendations.
Proofloop operates in two modes:
- Auto Mode: Hooks into skill/agent lifecycle events (e.g.
Stop) and automatically evaluates every execution. No user intervention required. Scores are persisted toskills/judge/scores/for trend analysis. - Manual Mode: Triggered explicitly via the
/judgecommand. The user specifies a skill name and optionally a transcript path. Useful for on-demand evaluation, re-scoring, or benchmarking.
Both modes produce the same structured scorecard output.
Scoring Dimensions
Proofloop evaluates across 7 dimensions. Each dimension receives a score from 1.0 to 10.0. The weighted composite determines the final grade.
| # | Dimension | Weight | What It Measures |
|---|---|---|---|
| 1 | Correctness | 25% | Output is factually correct. Code compiles and runs. No logical errors or bugs. |
| 2 | Completeness | 20% | All requirements from the prompt/task are addressed. Nothing is missing or skipped. |
| 3 | Adherence | 15% | The skill/agent followed its own SKILL.md or agent definition instructions precisely. |
| 4 | Actionability | 15% | Output is immediately usable without further manual work, fixes, or interpretation. |
| 5 | Efficiency | 10% | Minimal token waste. Appropriate tool usage. No unnecessary steps or redundant calls. |
| 6 | Safety | 10% | No harmful outputs. No data leaks. No destructive or irreversible actions taken without confirmation. |
| 7 | Consistency | 5% | Quality matches or exceeds previous executions of the same skill/agent. |
What ships with it
35 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- adapters/__init__.py 2.1 KB runs code
- adapters/claude_code.py 12 KB runs code
- adapters/codex.py 1.1 KB runs code
- adapters/cowork.py 663 B runs code
- adapters/openai_compatible.py 3.3 KB runs code
- analyzers/__init__.py 557 B runs code
- analyzers/llm_judge.py 22 KB runs code
- references/benchmark-standards.md 11 KB
- references/scoring-methodology.md 6.5 KB
- references/sycophancy_probes.json 2.9 KB
- rubrics/code-review.md 6.4 KB
- rubrics/content-writing.md 6.4 KB
- rubrics/custom-template.md 6.4 KB
- rubrics/data-analysis.md 7.0 KB
- rubrics/default.md 6.8 KB
- rubrics/devops.md 6.0 KB
- rubrics/documentation.md 6.6 KB
- rubrics/frontend-design.md 6.3 KB
- rubrics/research.md 6.0 KB
- rubrics/security.md 7.0 KB
- rubrics/security.weights.json 159 B
- rubrics/testing.md 6.5 KB
- scores/.gitkeep 0 B
- scripts/against.py 5.0 KB runs code
- scripts/benchmark.py 16 KB runs code
- scripts/compare.py 8.1 KB runs code
- scripts/cost_estimator.py 8.9 KB runs code
- scripts/detect-skill.sh 586 B runs code
- scripts/explain.py 26 KB runs code
- scripts/hook_lint.py 8.2 KB runs code
- scripts/report.py 11 KB runs code
- scripts/score.py 100 KB runs code
- scripts/studio.py 8.7 KB runs code
- scripts/watch.py 6.9 KB runs code
- SKILL-judge-explain.md 3.7 KB
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.
- 4d ago First seen · 382 lines · 22 tokens per session scan A 5478cb7821a5
judge is a skill published in the GitHub repository sattyamjjain/proofloop (5 stars, last pushed 2mo ago), licensed MIT. It adds 22 tokens to every session and 4,292 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.
Other skills, from other repositories
llm-evaluator
Evaluate LLM outputs systematically using LLM-as-judge, human evaluation frameworks, and regression testing. Use when assessing model quality, comparing models, or preventing quality regression.
assess
Assesses and rates quality 0-10 across multiple dimensions (correctness, maintainability, security, performance, testability, simplicity) with pros/cons analysis. Compares against project conventions and prior decisions from memory. Produces structured evaluation reports with actionable improvement suggestions. Use…
golden-dataset
Golden dataset lifecycle patterns for curation, versioning, quality validation, and CI integration. Use when building evaluation datasets, managing dataset versions, validating quality scores, or integrating golden tests into pipelines.
layerlens
Evaluate AI outputs with LayerLens. Upload traces, create judges, run evaluations, and retrieve quality scores -- all from within OpenClaw.
rag-eval
NVIDIA RAG Blueprint evaluation guidance for measuring retrieval and answer quality with stable datasets, baselines, and reproducible scoring workflows.
inno-idea-eval
Multi-persona idea evaluation with quality gate.