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 skills add fafa-ai-data-lab/ai-data-analyst-agent --skill data-validatorgit clone --depth 1 https://github.com/fafa-ai-data-lab/ai-data-analyst-agentWrote 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/fafa-ai-data-lab/ai-data-analyst-agent/data-validator)<a href="https://agentmods.dev/skills/fafa-ai-data-lab/ai-data-analyst-agent/data-validator"><img src="https://agentmods.dev/badge/skills/fafa-ai-data-lab/ai-data-analyst-agent/data-validator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/fafa-ai-data-lab/ai-data-analyst-agent/data-validator"><img src="https://agentmods.dev/badge/skills/fafa-ai-data-lab/ai-data-analyst-agent/data-validator.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00112 | $0.02676 |
| Opus 5.5 | $0.00045 | $0.01070 |
| Sonnet 5.5 | $0.00022 | $0.00535 |
| Haiku 4.5 | $0.00011 | $0.00268 |
Grade A, and why
data-validator 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 1mo 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 — 207 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Data Validator
独立于主流程的数据校验,避免自评偏差。自评不可信,校验必须分离(对应 Anthropic harness 设计原则:让生产者评判自己的产物有系统性偏差,验证必须交给独立 agent)。
关键约束:这个 agent 没有 Write / Edit 权限(见上方
allowed-tools)。它只能 Read / Grep / Bash(只读) / Glob,物理上无法改你的文件——它的唯一产出是一份PASS / WARNING / FAIL校验报告。
校验规则按主题分文件存放在 rules/,按场景按需加载。每被纠正一个错误 → 追加一条规则,下次自动生效。
校验场景(按需加载规则文件)
1. SQL 校验(写完 SQL、提交前)
Step 1:识别 SQL 类型并加载规则
- AB 实验 SQL(含实验分组逻辑)→ 加载
rules/abtest-rules.md - 涉及收入 / 订阅 / 付费 / 业务核心指标 → 加载你自己的
rules/<domain>-metric-rules.md(按_TEMPLATE.md建) - 所有 SQL → 加载你自己的
rules/<your>-field-traps.md(字段坑库)
Step 2:validator 用 Grep/regex 逐条 R-* 检查 SQL 文本,命中就报警。
Step 3:补充检查(规则不覆盖的)
- Pre-flight:未知表/字段是否已 DESCRIBE 确认
- 类型匹配:WHERE/JOIN 条件两侧类型是否一致
Step 4(关键):覆盖盲区检测 —— 见下文「## 覆盖盲区检测」。
2. 报表数据校验(数据拉完、写入报表/文档前)
加载:你的业务口径规则 + 文档格式规则(如有)。
规则不覆盖的额外检查:
- 环比合理性(如 ±30% 预警)
- 完整性(NULL / 0 / 缺失)
- 节假日 / 异常窗口过滤
- 跨源一致性(vs 已有看板 / 上一期)
3. 分析报告校验(写完、交付前)
加载:rules/report-style.md(必) + rules/sanity-check.md 的数字 grounding 规则。
校验方式:grep 报告 markdown 文本,逐条 R-RPT-* 检查。emoji 单独正则扫一遍。
4. SQL 结果合理性校验(SQL 跑完、用结果之前)
加载:rules/sanity-check.md(必)。
触发时机:
- 跑完 SQL 拿到结果,用结果之前
- 写完报告,交付之前
输入:SQL 文本 + 结果摘要(行数、关键聚合值、Top N 行)+ 业务上下文(预期粒度、对比基线)。
校验方式:validator 逐条 R-SAN-* 检查(结果空 / JOIN 膨胀 / 缺过滤 / 数字溯源)。
与场景 1 的关系:场景 1 是写 SQL 时的预防,场景 4 是跑完后的兜底,互补不重复。
5. ETL 修复回归校验(数据回刷完成后)
加载:你的 rules/known-bug-regression.md(按 _TEMPLATE.md 自建)。
机制:每修一个 ETL bug,agent 自决追加一条回归规则,含:bug 描述(根因 + 触发条件)、修复版本/日期、影响表、case、4 类回归 SQL(分布 / 业务规则 / 异常值 / 守恒)。长期形成 known-pattern 库。
这个文件高度业务相关(含真实表名、case),不要开源;本仓库只提供机制说明和模板。
6. 聚合/分摊表守恒对账(建表 / 改口径时)
加载:rules/sanity-check.md 的 R-SAN-007(必)。
核心:任何把上游度量(成本/收入/用量/计数)分摊或聚合到新粒度的表,必须 SUM(加工后) ≈ SUM(上游),差异 < 0.5%。单看加工表自身永远发现不了「少算/错配」,必须对上游总账。
覆盖盲区检测(每次必做)
校验完已知规则后,validator 必须主动识别「规则库没覆盖的新场景」,让人评估是否要沉淀新规则。否则规则库永远停留在已踩过的坑,新坑还是会被踩。
反幻觉铁律
涉及「与权威口径对比」的盲区,必须实际 Read 权威源文件验证,禁止凭印象判断,否则会 false positive 误报。涉及「表/字段是否文档化」的盲区,靠 grep 规则文件判断即可。
What ships with it
7 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.
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.
- 1mo ago First seen · 207 lines · 112 tokens per session scan A 6666524eecbe
data-validator is a skill published in the GitHub repository fafa-ai-data-lab/ai-data-analyst-agent (45 stars, last pushed 3mo ago), licensed MIT. It adds 112 tokens to every session and 2,676 once invoked, about $0.0004 per session on Opus 5.5. 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
clawhub
Dingo: A Comprehensive AI Data, Model and Application Quality Evaluation Tool.
fetch-data
A workflow that turns a plain-language request for business data into a database query, runs it in a data warehouse or lake, and returns the results for analysis.
bi-causal-attribution
A business-analysis skill that links business events in reports, plans, release notes, or conversations with an unusual metric change. It produces and ranks possible causes supported by the available evidence.
bi-clustering
A business-analysis skill that places similar business objects—such as users, products, or shops—into groups. It can use a two-by-two matrix based on two measures or clustering based on several features.
data-science-analysis
Computes a numeric or categorical answer to a quantitative data-science question by cleaning and analyzing local data files (CSV, Excel, TSV, and scientific formats .npz/.fits/.h5) with pandas, numpy, and scipy. Use whenever a task ships its own dataset (in whatever local directory it provides) and asks you to derive…
bi-report-generation
A process for turning completed business-data analysis into a readable HTML report with charts, tables, summaries, and source data.