Write a new LLM Gateway changelog entry. Use when the user says "changelog", "changelog entry", "write a changelog", "add a changelog", "announce this feature", or asks to document a shipped feature for the public changelog. Produces the dated markdown file under apps/ui/src/content/changelog plus a gpt-image-2 prompt for the OpenGraph image.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
changelog
description
Write a new LLM Gateway changelog entry. Use when the user says "changelog", "changelog entry", "write a changelog", "add a changelog", "announce this feature", or asks to document a shipped feature for the public changelog. Produces the dated markdown file under apps/ui/src/content/changelog plus a gpt-image-2 prompt for the OpenGraph image.
Changelog
Write a public changelog entry for LLM Gateway, in the house style, and hand back a ready-to-run gpt-image-2 prompt for its OpenGraph image.
What you need first
Before writing, make sure you understand the feature concretely. If the user only gave a feature name, find the facts — don't guess:
Read the relevant docs page under apps/docs/content/ (e.g. features/<feature>.mdx).
Inspect the shipping commit/PR if one is referenced: git show <sha> --stat, then read the changed UI/API/gateway files for the exact user-facing behavior.
Confirm plan gating (free vs Pro vs Enterprise), exact field names, error codes, and any limits. The changelog must be accurate — never invent prices, limits, or capabilities.
If this prints an existing duplicate, report it and still choose an unused id
above the current maximum; do not silently renumber published entries.
slug: short kebab-case, feature-focused (e.g. custom-model-catalog). The slug must match the filename suffix and the image.src filename, and becomes the URL /changelog/<slug>.
Step 2 — Write the markdown file
Frontmatter (YAML), then the body. Entries are plain Markdown — no MDX/JSX components (no <Callout>); use bold inline notes instead.
---
id: "<next-id>"
slug: "<slug>"
date: "<YYYY-MM-DD>"
title: "<TitleCase, ~3–7words>"
summary: "<1–3 sentences: what shipped, the concrete benefit, and the plan if gated. This is the OG description and the listing blurb.>"
image:
src: "/changelog/<slug>.png"
alt: "<Descriptivealttext:thefeatureandwhattheimageshows>"
width: 1536
height: 1024
---<Openingparagraph:theproblemthissolves, thentheone-linestatementofwhatshipped.Boldthefeaturenameonce.>## <Sectionheader — verb-ledoroutcome-led><Body.Preferatableforfield/optionreferencesandafencedcodeblockforanAPIexample.>
---
**[<Docs link> →](https://docs.llmgateway.io/...)** | **[<Secondary CTA> →](https://llmgateway.io/...)**
House style (match existing entries)
Read the two or three most recent files in apps/ui/src/content/changelog/ before writing, and mirror their tone:
Lead with the problem, then the fix. First paragraph names the pain; the next sentence states what shipped.
Benefits over features, specific over vague. "about 50% cheaper", "rejected with 400 before any data leaves the gateway" — not "improved efficiency".
Confident and plain. Active voice. No exclamation points. No "very/really/simply". No buzzwords ("streamline", "seamless", "revolutionary").
Show the API. Include a realistic curl or JSON example when there's a request-level change. Use https://api.llmgateway.io/v1/... and $LLM_GATEWAY_API_KEY.
Use a table for fields, options, strategies, or tiers.
State plan gating explicitly (e.g. "Available on the Enterprise plan").
Close with a footer link line: a verified docs link and one verified
secondary CTA, separated by | and bolded.
Keep section headers in ##. Keep it scannable — short paragraphs, bullets for lists of behaviors.
Step 3 — Produce the OpenGraph image prompt
Hand the user a single, ready-to-paste gpt-image-2 prompt that produces the OG image, plus where to save it.
Resolution. Target 1536×1024 (3:2 landscape), matching current
changelog entries and the repository's gpt-image-2 image presets. Match the
frontmatter dimensions to the generated file.
Prompt guidance. Write a small summary prompt — 2–4 sentences — in the house image style: a glossy 3D-rendered circuit-board scene, not a flat/minimal gradient backdrop. The prompt should:
Set the scene: a dark navy computer circuit board in glossy 3D isometric perspective, with bright neon-teal light traces flowing across it toward a central raised chip.
Put the feature's concept at the center: a glowing element mounted on the central chip (e.g. a glowing doorway for the gateway, a glowing key for API keys). Concept over literalism.
Surround it with supporting glossy rounded 3D icons that fit the feature (chat bubbles, keys, charts, coins…) in vivid purple, lime green, and mint, each on small pedestals on the board.
Add the render feel: subtle depth of field at the edges, soft reflections, premium 3D render, vibrant against the dark board.
Says "no text, no words, no letters, no UI chrome" — AI image text is unreliable; the title lives in the page, not the image.
Specifies the aspect: "wide 3:2 landscape composition, 1536×1024".
Output the prompt in a fenced block, then the save path, e.g.:
Prompt for gpt-image-2 (1536×1024):
```
A dark navy computer circuit board in glossy 3D isometric perspective, with
bright neon-teal light traces flowing across it toward a central raised chip.
On the chip sits a glowing <concept element>. Around it, glossy rounded 3D
<supporting icons> in vivid purple, lime green, and mint stand on small
pedestals on the board. Subtle depth of field at the edges, soft reflections,
premium 3D render. Wide 3:2 landscape composition, 1536×1024. No text, no
words, no letters, no UI chrome.
```
Save the result to: apps/ui/public/changelog/<slug>.png
Generate the image with gpt-image-2, then put the PNG at that path and inspect
its actual dimensions before validating.
Step 4 — Validate
pnpm format
pnpm exec turbo run build --filter=ui
pnpm format normalizes the markdown; the ui build fails if the frontmatter doesn't match the content-collections schema. Then commit (conventional commit, ≤50-char title), e.g. docs(changelog): add custom model catalog entry.
