菜单
Ingestion patterns for CSV, JSON, REST API, SQL, Excel, and Parquet -- guides an LLM through loading data from any common source
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/fabioc-aloha/Alex_Skill_Mall --skill datasource-connectors命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | datasource-connectors |
| description | Ingestion patterns for CSV, JSON, REST API, SQL, Excel, and Parquet -- guides an LLM through loading data from any common source |
| tier | standard |
| applyTo | **/*data*,**/*ingest*,**/*connect*,**/*csv*,**/*json*,**/*sql*,**/*excel*,**/*parquet*,**/*api* |
| currency | 2026-05-02T00:00:00.000Z |
| lastReviewed | 2026-05-02T00:00:00.000Z |
Guide the user through loading data from any common source into a working dataset. This module is a decision framework, not a library. It tells the LLM what to ask, what to watch for, and how to handle each format's quirks.
Ask the user what format the data is in, or infer from the path/URL:
| Signal | Connector |
|---|---|
.csv, .tsv, .txt (tabular) | CSV |
.json, .jsonl, .ndjson | JSON |
http://, https:// + returns JSON/XML | REST API |
Connection string, .sql, database name | SQL |
.xlsx, .xls | Excel |
.parquet, .arrow | Parquet |
If ambiguous, ask. Do not guess the format.
| Problem | Detection | Fix |
|---|---|---|
| Wrong delimiter | First row has one column | Try tab, semicolon, pipe |
| Encoding garbled | Non-ASCII chars show as ? or é | Re-read as Latin-1 or Windows-1252 |
| Trailing commas | Row has one extra empty column | Strip trailing delimiter |
| Quoted fields with commas | Standard CSV; most parsers handle it | Use RFC 4180 compliant parser |
| Mixed line endings | \r\n and \n in same file | Normalize to \n |
| BOM marker | \xEF\xBB\xBF at file start | Strip or let parser handle |
A table with typed columns. Report: row count, column count, detected delimiter, detected encoding.
data.results[], root array)| Problem | Detection | Fix |
|---|---|---|
| Nested objects | Column values are objects, not scalars | Flatten with dot notation (address.city) |
| Mixed types in array | Some items have extra/missing keys | Union all keys; fill missing with null |
| Large file (> 100MB) | Slow to parse | Stream with JSONL or chunk |
| Date strings | ISO 8601 or Unix timestamps | Parse to date; state the format detected |
A flat table. Nested objects flattened to dot-notation columns. Report: record count, nesting depth, keys found.
results[])| Method | How to use | Security |
|---|---|---|
| API key in header | Authorization: Api-Key <key> or custom header | Never log the key |
| Bearer token | Authorization: Bearer <token> | Store in env var or secret |
| OAuth 2.0 | Client credentials or authorization code flow | Use refresh tokens |
| No auth | Public API | Still rate-limit respectfully |
Never hardcode credentials. Use environment variables or secret storage.
| Pattern | Detection | Handling |
|---|---|---|
| Offset | ?offset=0&limit=100 | Increment offset until empty page |
| Cursor | Response has next_cursor field | Pass cursor to next request |
| Link header | Link: <url>; rel="next" | Follow rel="next" until absent |
| Page number | ?page=1&per_page=50 | Increment page until empty |
Set a maximum page limit (default: 100 pages) to prevent runaway loops.
| Rule | Value |
|---|---|
| Timeout per request | 30 seconds |
| Retries on 429/5xx | 3 with exponential backoff (1s, 2s, 4s) |
| Retries on network error | 2 |
| Never retry | 400, 401, 403, 404 (client errors are not transient) |
A flat table from all pages combined. Report: total records, pages fetched, any errors skipped.
| Rule | Reason |
|---|---|
| Always use parameterized queries | Prevents SQL injection |
| Never interpolate user input into SQL | Even for table names; use allowlists |
| Read-only connection | The pipeline only reads; never write |
| Credentials in env vars | Never in code, never in logs |
-- Simple table read
SELECT * FROM schema.table_name LIMIT 1000;
-- Filtered read (parameterized)
SELECT * FROM sales WHERE region = @region AND date >= @start_date;
-- Aggregated read (push aggregation to the database when possible)
SELECT region, month, SUM(revenue) AS revenue
FROM sales
GROUP BY region, month;
Push aggregation and filtering to the database when possible. Pulling raw data and aggregating in the LLM wastes tokens and is slower.
A table. Report: row count, column count, query execution time, database engine.
| Problem | Detection | Fix |
|---|---|---|
| Multiple sheets | Workbook has > 1 sheet | Ask which sheet; default to first |
| Header not in row 1 | First rows are title/metadata | Ask for header row number |
| Merged cells | Values span multiple rows/columns | Unmerge and fill down |
| Formulas | Cells contain formulas, not values | Read values, not formulas |
| Mixed types in column | Numbers and text in same column | Coerce to string; flag for data-preparation |
| Date serial numbers | Dates stored as integers (e.g., 45678) | Convert using Excel epoch (1899-12-30) |
A table from the specified sheet and range. Report: sheet name, row count, column count, any merged cells detected.
Parquet is self-describing: schema, types, and encoding are embedded. No delimiter guessing, no encoding issues, no header detection.
| Problem | Detection | Fix |
|---|---|---|
| Partitioned directory | Path is a directory with part-*.parquet files | Read all partitions |
| Nested columns (struct/list) | Schema shows complex types | Flatten or ask user which fields |
| Large file | > 1M rows | Sample or filter before loading all |
A typed table. Report: row count, column count, schema summary (types from the Parquet metadata).
| Situation | Action |
|---|---|
| File not found | Report the exact path tried. Ask user to verify. |
| Permission denied | Report the error. Do not retry. Ask user to fix permissions. |
| Empty dataset (0 rows) | Report it. Push back: "file exists but has no data rows." |
| Partial data (truncated) | Report row count. Ask if this is expected. |
| Unsupported format | State what was detected. Ask user for the correct format. |
Never fail silently. Every error must produce a specific, actionable message.
Pass to the next module (data-preparation):