本文へ移動
cccskills
無料GitHub で公開

doris-img-gen

Create Apache Doris article illustrations in the Doris hand-drawn whiteboard style, using a randomly selected Pip, Dori, or Flux mascot only when a task, guide, operator, diagnosis, or anthropomorphic narrative benefits from a character. Use when the user invokes $doris-img-gen or asks for a Doris technical article illustration, concept visual, infographic, architecture diagram, process diagram, or mascot-assisted explanatory graphic.

日本語の概要は準備中です。原文の説明を表示しています。

apache/doris-websiteスター 136(GitHub 136 + cccskills 0)

実行スコア 未検証

インストール方法を見る

含まれるファイル(5)

  • SKILL.md6.7 KB
  • agents/openai.yaml262 B
  • assets/meet-d-crew.jpg161.0 KB
  • references/visual-style.md2.9 KB
  • scripts/select_mascot.py1.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Doris Article Illustration Generator

Create technically clear, visually consistent illustrations for Apache Doris articles. Treat the text after $doris-img-gen as the primary content brief.

Prepare

  1. Read references/visual-style.md completely before generating an image.
  2. Extract the requested concepts, relationships, sequence, labels, and layout constraints. Preserve them; do not invent technical facts, metrics, components, or product claims.
  3. If the request refers to an article, section, or local draft, read the relevant passage before composing the illustration.
  4. Keep all in-image wording in English by default. Use the user's exact quoted labels when supplied. Keep labels short and render each label exactly once.

Decide Whether to Use a Mascot

Apply the semantic-role test: include a mascot only when it performs a meaningful action that can be stated as a verb in the image.

Include one mascot when the scene needs:

  • an operator starting or running a task;
  • an investigator searching, observing, comparing, or diagnosing;
  • a guide teaching or walking readers through a process;
  • a central actor already present in the requested narrative;
  • a human-like moment that makes an abstract idea easier to understand.

Omit the mascot when the image is primarily:

  • architecture, topology, data flow, or component relationships;
  • a chart, table, query plan, dashboard, or factual comparison;
  • a compact process whose clarity would be reduced by a character;
  • a scene where the mascot would be decorative rather than explanatory.

Never add a mascot only to fill space. Use one mascot by default. Use more than one only when interaction among roles is itself part of the requested content.

An explicit user request for a character always overrides random selection.

Treat the character roles as personality and prop hints, not fixed assignment rules:

  • Pip — The Query Runner: purple triangular character; can run, launch, point, inspect, or guide.
  • Dori — The Core Engine: cyan capsule character; can compute, operate, inspect, compare, or guide.
  • Flux — The Data Explorer: mint-green diagonal capsule character; can search, investigate, connect, inspect, or guide.

Randomize Mascot Selection

After applying the semantic-role test, count only the images that need a mascot and do not already have a user-specified character. Randomize those slots by running the bundled selector once from the skill directory:

python3 scripts/select_mascot.py --count <number-of-unspecified-mascot-slots>

Assign the returned names to the mascot-bearing images in output order. For a single image, omit --count or pass --count 1.

The selector creates a fresh system-random order on every invocation. In a batch, it uses every character once before reshuffling and avoids the same character on adjacent slots. Use its output exactly: do not replace or rerun a valid result because another mascot seems more suitable for the scene. Do not run the selector for images that fail the semantic-role test.

Inspect assets/meet-d-crew.jpg with an image-viewing tool whenever a mascot will appear. Use it as the identity reference and preserve the selected character's silhouette, body color, face, and simple black limbs. Translate the character into the illustration style instead of copying the reference card layout or its surrounding text.

Compose the Image Prompt

Build a structured prompt with these fields:

Use case: infographic-diagram
Asset type: Apache Doris technical article illustration
Primary request: <content and relationships from the user's brief>
Scene/backdrop: <only what is needed to explain the content>
Subject: <main concepts, icons, arrows, and the mascot's action if included>
Style/medium: <the canonical style from references/visual-style.md>
Composition/framing: landscape 1200 x 630 with a clear reading order and generous margins
Text (verbatim): <exact English labels, or "No text">
Color palette: <the canonical palette from references/visual-style.md>
Constraints: technically faithful; readable labels; consistent icon scale; clear arrow direction; no extra text; no watermark; no photorealism; no 3D
Avoid: decorative clutter; dense paragraphs; tiny labels; unrelated logos; mascot use without a semantic role

If using the D-Crew reference, add:

Input images: Image N is the D-Crew identity reference. Use only the selected mascot; do not reproduce the reference poster, logo lockup, cards, headings, or descriptive copy.
Character consistency: preserve the selected mascot's recognizable geometry, color, face, and limbs; render it with the same hand-drawn marker linework as the rest of the illustration.

If the user also supplies a composition reference, identify each input image by number and role. Preserve the composition's information architecture without copying irrelevant artifacts.

Generate and Review

  1. Use the built-in image-generation tool by default.
  2. For a new image without a mascot or other supplied image reference, omit all reference-image parameters.
  3. When using a local mascot or composition reference, pass the required local files through referenced_image_paths and describe each image's role explicitly.
  4. Save the final image in the user's project or requested destination as a PNG with a descriptive filename.
  5. Inspect the result at full resolution. Verify:
    • all requested concepts and relationships are present;
    • arrow directions and process order are correct;
    • every label is English, legible, and spelled correctly;
    • no unrequested wording appears;
    • the mascot decision adds meaning, and the chosen character follows the random rotation or the user's explicit request;
    • the image follows the canonical palette, background, linework, and dimensions.
  6. If any verification fails, refine the smallest necessary area and inspect again.

Invocation Examples

$doris-img-gen Show a query diagnosis loop with the exact labels "Hypothesis", "Evidence", "Observation", and "Elimination". A data investigator reviews profile, plan, chart, and checklist evidence.
$doris-img-gen Draw a left-to-right Doris data flow from client to FE to BE to object storage. Use no mascot and no text beyond the component labels.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

コメント

ログインなしで書き込めます。使ってみた感想、質問、動かなかった環境などを気軽にどうぞ。

まだコメントはありません。最初のコメントを書いてみませんか。

送信するとすぐに公開されます。メールアドレスや電話番号などの個人情報は書かないでください。URL は 2 つまでです。

0 / 1000 文字

このフォームは reCAPTCHA で保護されており、Google のプライバシーポリシーと利用規約が適用されます。

同じリポジトリのスキル

Publish a new Apache Doris Core or Doris ecosystem release on the Doris website. Use when updating download data, bilingual release notes, the Doris Core release index, release navigation, or ecosystem release pages. Confirm the component and release metadata first because Core and ecosystem releases use different files and validation paths.

日本語の概要は準備中です。原文の説明を表示しています。

apache/doris-website1362026年10月29日 更新

Restructure and SEO/GEO-optimize an existing Apache Doris user documentation file (`.md` / `.mdx`), primarily under `i18n/zh-CN/docusaurus-plugin-content-docs-next/current/` or `docs-next/`. The skill reorganizes the doc from a user-scenario perspective, applies tables and lists for parameters / scenarios / FAQs, fixes formatting (4-space indent, code-block language tags, blank lines, Chinese/English spacing, command-concatenation bugs), validates every external link with `curl` and every `/images/...` reference against `static/`, and updates the frontmatter `description` + `keywords` per the bundled `./references/seo-geo.md` guide — all without changing the original meaning or dropping any technical content (commands, parameters, YAML, sample outputs, images are preserved verbatim). Use this skill whenever the user points at a single Doris doc file and asks to "优化这篇文档", "重新组织这篇文档", "帮我优化文档结构", "对这篇文章进行 SEO 优化", "按 SEO/GEO 优化这篇文档", "调整一下这篇文档的结构", "把这篇文档润色一下", "optimize this doc", "restructure this doc", "polish this doc", or anything similar — even if the user only says "make this doc better" or "这篇文档读起来不顺" while pointing at a Doris doc path. Do NOT use this skill for translating between languages (use `doris-translate-zh-to-en` instead), writing a brand-new doc from scratch (use `doris-feature-card` or `doc-coauthoring`), or pure link-checking without restructure (use `check-md-links` instead).

日本語の概要は準備中です。原文の説明を表示しています。

apache/doris-website1362026年10月29日 更新

Audit and update the user documentation for an Apache Doris release from the commit diff between two refs (for example 4.1.3..4.1.4-rc04). Extract the hard user-facing surface (FE/BE/MS configs, session variables, grammar, functions, system tables, HTTP endpoints, metrics, shipped conf files, dependency versions), classify every commit with parallel subagents, then add, fix or create docs under three iron rules — anything a user can perceive needs a doc, every new feature or behavior change carries a "since version X.Y.Z" note, and every feature commit needs a documented home or a question to the user. Write the Chinese version of ONE doc branch first (ask which one: dev or 4.x), open a PR, and only after the user approves it sync to the other branches and to English (dev documents master, so build a master-vs-tag difference table first). Use when asked to "check the docs for 4.1.5", "update docs from the commit history", "what docs need to change for X.Y.Z", or "sync the 4.x doc changes to dev and English". Not for writing release notes and not for building the site (static checks only).

日本語の概要は準備中です。原文の説明を表示しています。

apache/doris-website1362026年10月29日 更新

Generate an English Slack-ready post and save a temporary Markdown copy from an Apache Doris weekly community report MDX file. Use when the user invokes `$weekly-report-slack path/to/report.mdx` or asks to turn a Doris weekly report, especially `src/pages/community-report/_reports/YYYY-MM-DD.mdx`, into the standard Slack announcement format with date range, summary paragraph, highlights, and report link.

日本語の概要は準備中です。原文の説明を表示しています。

apache/doris-website1362026年10月29日 更新

apache のスキルをすべて見る

このスキルの問題を報告する