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/ryanzhao1011/workframe/prompt-evaluationnpx skills add ryanzhao1011/workframe --skill prompt-evaluationgit clone --depth 1 https://github.com/ryanzhao1011/workframeWrote 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/ryanzhao1011/workframe/prompt-evaluation)<a href="https://agentmods.dev/skills/ryanzhao1011/workframe/prompt-evaluation"><img src="https://agentmods.dev/badge/skills/ryanzhao1011/workframe/prompt-evaluation.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 | $0.00039 | $0.01682 |
| Opus 5 | $0.00019 | $0.00841 |
| Sonnet 5 | $0.00008 | $0.00336 |
| Haiku 4.5 | $0.00004 | $0.00168 |
Grade A, and why
prompt-evaluation 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 — 170 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Prompt 评估技能
适用场景
- 评估单个 Prompt 的输出质量
- 对比多个候选 Prompt，选出最优
- 验证 Prompt 迭代后的效果（前后对比）
- 定期回归验证，防止 Prompt 效果衰减
五步流程
第 1 步：评估维度定义
根据 Prompt 场景选择评估维度（可叠加）：
| 维度 | 评估问题 | 适用场景 |
|---|---|---|
| 准确性 | 输出内容是否事实正确、任务完成度高？ | 信息类、生成类 |
| 完整性 | 是否覆盖了所有要求的要素？ | 结构化输出 |
| 一致性 | 相同输入是否产出相似输出？ | 稳定性要求高的场景 |
| 可读性 | 输出是否清晰易懂、格式合理？ | 面向终端用户 |
| 相关性 | 输出是否贴合用户意图？ | 问答、推荐 |
| 安全性 | 是否规避了违规/有害/越权内容？ | 全部 |
| 成本 | 消耗的 token 数、延迟 | 高频调用场景 |
| 鲁棒性 | 对边界/对抗输入的处理能力 | 高安全要求 |
每个维度定义 1-5 分的评分标准。
第 2 步：测试样本设计
设计代表性测试样本集：
| 样本类别 | 占比 | 用途 |
|---|---|---|
| 正常样本（Typical） | 50% | 覆盖主流场景 |
| 边界样本（Edge） | 30% | 测试边界处理 |
| 对抗样本（Adversarial） | 20% | 测试安全性和鲁棒性 |
样本集规模建议：
- 快速验证：10-20 个样本
- 标准评估：30-50 个样本
- 深度回归：100+ 个样本
每个样本包含：
sample_id: S-001
category: typical | edge | adversarial
input:
variables:
user_input: "{具体输入}"
style_tone: "neutral"
expected_characteristics:
- "应包含 X"
- "不应包含 Y"
- "格式应符合 Z"
第 3 步：对比测试执行
| 对比模式 | 说明 |
|---|---|
| 单 Prompt 评估 | 跑一个 Prompt，打分判断是否达标 |
| A/B 对比 | 两个候选 Prompt 在同一样本集上对比 |
| 版本回归 | 新版 Prompt vs 旧版 Prompt，防止退步 |
| 多候选排序 | 3+ 个候选 Prompt 排序选最优 |
执行时注意：
- 固定 temperature / top_p 等参数，保证可复现
- 每个样本至少运行 3 次（减少随机性影响）
- 记录完整输入/输出/耗时/token
第 4 步：量化评分
按维度为每个输出打分（1-5），计算综合得分：
sample_id: S-001
prompt_version: v1.2
scores:
accuracy: 4
completeness: 5
consistency: 4
readability: 5
safety: 5
overall: 4.6 # 加权平均或简单均值
cost:
input_tokens: 150
output_tokens: 320
latency_ms: 1200
汇总到评估矩阵：
| 样本 | Prompt A | Prompt B | 胜出 |
|---|---|---|---|
| S-001 | 4.6 | 4.2 | A |
| S-002 | 3.8 | 4.5 | B |
| ... | ... | ... | ... |
| 平均 | 4.2 | 4.1 | A |
第 5 步：上线建议
基于评分给出明确建议：
| 结果 | 建议 | 后续动作 |
|---|---|---|
| 新版显著优于旧版（>10%） | ✅ 上线 | 灰度发布 → 全量 |
| 新版略优（2-10%） | ⚠️ 谨慎上线 | 小流量 A/B 验证 |
| 效果持平（±2%） | 🔄 不上线 | 继续优化或保持现状 |
| 新版变差 | ❌ 回退 | 分析原因，重新设计 |
| 某维度显著回退 | ❌ 回退 | 即使综合分更高也回退 |
输出模板
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 · 170 lines · 39 tokens per session scan A f47079caeaa5
prompt-evaluation is a skill published in the GitHub repository ryanzhao1011/workframe (4 stars, last pushed 18d ago), licensed MIT. It adds 39 tokens to every session and 1,682 once invoked, about $0.0002 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
optimize
Rephrase a rough prompt to follow prompt-engineering best practices WITHOUT changing its meaning, then show it for review without executing. Use when the user runs /petprompt:optimize or asks to rewrite/clean up/optimize their prompt before running it.
cursor-prompting
Prompt shaping rules for delegated cursor-agent tasks.
agent-orchestration-improve-agent
Systematic improvement of existing agents through performance analysis, prompt engineering, and continuous iteration.
prompt-cookbook
Build a prompt cookbook for one company's actual vertical and roles rather than generic examples. Each recipe names the job it does, who runs it, the prompt itself, what good output looks like, and how to tell when it went wrong. Written for people who have never written a prompt and will not read documentation about…
skill-meta-prompt
Craft better prompts using proven optimization techniques — use when your prompt needs refinement.
ax-gen
This skill helps an LLM generate correct AxGen code using @ax-llm/ax. Use when the user asks about ax(), AxGen, generators, forward(), streamingForward(), validation, assertions, streaming assertions, field processors, step hooks, self-tuning, or structured outputs. For MCP clients, transports, prompts, resources…