agentsclimarketplace

Dolphindb rag

Skill dolphindb/DolphinX_Skill/dolphindb-factor/dolphindb-rag

MUST use before answering any DolphinDB question — enforces source-grounded answers (no fabrication from memory) when no retrieval tool is available. Code execution happens on the DolphinDB.From its SKILL.md

Install
npx -y skills add dolphindb/DolphinX_Skill --skill dolphindb-rag

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.1 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

DolphinDB Answer Discipline

The Rule

NEVER answer DolphinDB questions from memory alone. Ground every claim in a verifiable source.

Model training data is outdated and frequently wrong about DolphinDB (function names, signatures, division operator, window semantics). Without a retrieval tool available, the only safe sources are:

  1. Official DolphinDB documentation the user has linked or pasted
  2. Local docs / scripts the user has in this workspace (read them with the file tools)
  3. Verified behavior from executing a minimal probe script on the DolphinX platform
  4. The schema / pitfalls notes shipped with dolphindb-daily-factor (references/schema.md, references/pitfalls.md)

Anything else is a guess. Mark it as a guess or refuse.

Scope of This Skill

This skill is answer discipline for DolphinDB questions:

  • ✅ Explaining DolphinDB syntax / functions when you can cite a concrete source
  • ✅ Diagnosing error messages by either matching them to local docs or running a reproducer on the DolphinX platform
  • ✅ Recommending a pattern, with the source the recommendation came from

Out of scope:

  • ❌ Generating daily-frequency factor code → use the dolphindb-daily-factor skill (it owns the template, schema, and execution loop)
  • ❌ Inventing a function signature because it "feels right" → STOP, say you're not sure and ask the user for the doc page, or run a probe on the platform

When in Doubt — Verify Before You Answer

In order of preference:

  1. Ask the user to paste / link the relevant doc section if they have it. One sentence: "I want to verify X — do you have the doc page for <function> handy?"
  2. Read local files — if the workspace has DolphinDB scripts, modules, or docs, search them with the file tools before answering.
  3. Probe the DolphinX platform — run the smallest possible script that reveals the answer (e.g. call the function on a tiny vector and inspect the result / error). Cheaper than wrong answers.
  4. Say "I'm not sure" — explicitly. Do not paper over uncertainty with confident prose.

If the question is about live cluster state (does a DB exist? what tables? what columns?), always go to option 3 — see dolphindb-daily-factor/references/probing.md for the catalog commands.

Workflow

For a typical DolphinDB question:

  1. Identify what needs verification — list the function names, operators, and behaviors the answer depends on.
  2. For each item, pick a source:
    • User-provided doc → cite it
    • Local file → read and cite path + line range
    • Probe script → run on the DolphinX platform, cite the observed output
    • None of the above → flag as unverified, ask the user or refuse
  3. Compose the answer from verified pieces only. No filler from memory.
  4. If the answer is a runnable script, the user can execute it directly on the DolphinX platform.

If the user reports an error message:

  • Match the verbatim error against local docs / scripts first
  • If no match, build a minimal reproducer and run it on the DolphinX platform, then diagnose from the observed behavior

Response Format

[Answer composed from verified sources only.]

来源 / Sources:
- <doc title or local file path> (L<start>-<end> if applicable)
- DolphinX 平台执行结果: <one-line summary of what was run and what came back>

If a part of the answer could not be verified, say so explicitly:

⚠️ 以下内容未在本会话中验证: <claim>
   建议: <how the user can verify — paste doc, run on platform, etc.>

Checklist (before sending the answer)

  • Every non-trivial claim has a source (user doc, local file, or platform-execution observation)
  • No fabricated function names, signatures, or behaviors
  • Unverified parts are flagged with ⚠️, not buried
  • If the answer is a factor / panel script, told the user to use dolphindb-daily-factor
  • Code snippets in the answer are minimal and the user can run them on DolphinX as-is

Red Flags — You're Doing It Wrong

  • "据我所知 DolphinDB..." / "DolphinDB should support..." without a source → STOP, verify or flag
  • Guessing the cause of an error message without reproducing it → STOP, build a minimal probe
  • Making up DolphinDB syntax from Python / SQL / Java intuition → STOP, DolphinDB syntax is different (\ vs / division being the canonical trap)
  • "Basic syntax is common knowledge" → WRONG, verify it
  • Confidently stating a function signature you have not seen this session → STOP
  • Skipping the source citation → STOP, every claim needs a source
  • Answering "does this database/table/field exist?" from memory → WRONG, run existsDatabase / getTables / schema on the DolphinX platform (see dolphindb-daily-factor/references/probing.md)
  • Trying to run / verify a generated factor script here → WRONG skill, hand off to dolphindb-daily-factor

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.