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/cass-2003/local-workflow-skill/llm-evalnpx skills add cass-2003/local-workflow-skill --skill llm-evalgit clone --depth 1 https://github.com/cass-2003/local-workflow-skillWrote 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/cass-2003/local-workflow-skill/llm-eval)<a href="https://agentmods.dev/skills/cass-2003/local-workflow-skill/llm-eval"><img src="https://agentmods.dev/badge/skills/cass-2003/local-workflow-skill/llm-eval.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 | $0.00057 | $0.01761 |
| Opus 5 | $0.00028 | $0.00881 |
| Sonnet 5 | $0.00011 | $0.00352 |
| Haiku 4.5 | $0.00006 | $0.00176 |
Grade A, and why
llm-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 5d 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 — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LLM Eval
LLM Eval 负责评估 LLM、Agent、RAG、prompt 和模型变更的效果与风险。目标是把“感觉更好”变成可复验的数据集、指标、评审标准、误差分析和上线门禁。
适用范围
- LLM、Agent、RAG、prompt、model eval、hallucination、faithfulness、tool-call accuracy 和 answer quality。
- judge prompt、golden dataset、rubric、human review、pairwise comparison、A/B 模型对比和回归集。
- RAG 检索命中、引用一致性、上下文利用、拒答、越权、工具调用、计划执行和多轮任务成功率。
- CI gate、离线评测、线上抽样、漂移监控、成本/延迟/质量 tradeoff 和发布评估报告。
不适用范围
- 普通软件单元测试、接口测试、E2E、覆盖率和回归测试;这些走 test-engineering。
- 普通 AI 工程实现、RAG 后端、向量库、Agent 工具开发和模型服务接入;这些走 ai-engineering。
- 纯 prompt 文案润色、营销文案或单次回答优化,没有评测数据和指标。
- 只读学习、项目上手、仅了解 LLM eval 概念,没有评估、设计、运行或上线门禁动作。
铁律
- Eval 必须先定义任务、用户、失败代价、基线、候选、数据集和上线阈值。
- 不能只靠一个 judge 分数;关键任务要结合规则指标、人工抽样、误差标签和案例复盘。
- Judge prompt 本身要评测:一致性、偏置、位置偏差、长度偏差、泄漏、可解释性和复判样本。
- RAG 评测必须拆开检索、重排、上下文利用、引用真实性和最终回答,不能只看总分。
- Tool-call eval 必须检查工具选择、参数、调用顺序、错误恢复、权限边界和最终状态。
- 数据集必须去重、分层、版本化;不能把训练/调参样本当最终验收集。
- A/B 对比必须固定输入、模型配置、温度、工具、知识库版本和评分口径。
- CI gate 要有可解释失败样本和豁免流程;不能让随机波动直接阻断所有发布。
强制流程
- 锁定目标:确认要评估的产品任务、模型/agent/RAG/prompt 版本、用户场景和失败代价。
- 建数据集:收集真实样本、合成边界样本、负例、权限样本、长尾样本和回归样本;记录来源和版本。
- 定指标:选择准确性、完整性、faithfulness、citation、tool accuracy、refusal、latency、cost、safety 和 UX 指标。
- 设计判分:为每个指标写 rubric、可接受答案、拒答规则、证据要求和人工复核策略。
- 运行评测:固定模型参数、工具版本、知识库版本、随机种子或重复次数;保存输入、输出、评分和日志。
- 误差分析:按 hallucination、retrieval miss、bad context use、wrong tool、bad parameter、policy miss、format fail 分类。
- 决策门禁:对比基线和候选,给出通过、回滚、继续调参或扩大人工评审的结论。
- 持续监控:把关键 eval 进 CI 或定期任务,定义漂移报警、样本补充和阈值调整。
场景执行卡
安全与门禁补强
- 安全 eval 必须覆盖 direct/indirect prompt injection、malicious retrieved docs、tool poisoning、data exfiltration、jailbreak、policy bypass 和越权工具调用。
- Gate 要写清最低样本量、重复次数、基线 delta、波动容忍、flaky 豁免、人工复核和回滚触发条件;不能只给一个平均分。
- 评测证据至少包含 run_id、dataset_version、prompt/model/tool/corpus/scorer version、运行时间、脱敏输入输出、trace/log 链接和失败样本 topN。
- 指标要可计算:tool exact match、schema validity、citation support、faithfulness、judge agreement、p95 latency、cost/sample 和 refusal accuracy。
- Judge 需要 gold set、防泄漏、inter-judge agreement、人工仲裁和边界样本复判;不能让单个 judge 自证正确。
- 调参集、开发集和最终验收集必须分离;若复用样本,必须标记污染风险并降低结论等级。
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.
- 5d ago First seen · 100 lines · 57 tokens per session scan A e4c7d1233dee
llm-eval is a skill published in the GitHub repository cass-2003/local-workflow-skill (12 stars, last pushed 1mo ago), licensed MIT. It adds 57 tokens to every session and 1,761 once invoked, about $0.0003 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 2026-08-30.
Other skills, from other repositories
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
google-cloud-solution-agentic-analytics-spark-knowledge-catalog
Discovers requirements and generates guidance to design and deploy a governed, secure agentic-analytics solution for data that's distributed across Google Cloud, other cloud providers, or on-premises. Data that's outside Google Cloud (such as data from Databricks, Snowflake, Salesforce, SAP, or Oracle systems) is…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.