article-poster

Name: article-poster
Author: MagicCube

489stars57forks

How to install article-poster?

magiccube/agentara/article-poster

$npx -y skills add MagicCube/agentara --skill article-poster

Installs into the current project.

Summary

Generate a beautifully designed infographic poster from an article URL or text content. Trigger when user says "article poster", "文章海报", "infographic", "信息图", "make a poster", "生成海报", "visual summary", or requests to convert an article/blog into a shareable image. NOT for generic

Files · 1

View on GitHub

SKILL.md

1	# Article Poster
2
3	Converts an article (URL or pasted text) into a beautifully designed infographic poster image.
4
5	## How It Works
6
7	1. Read the article — Fetch URL via `web_fetch` or use provided text
8	2. Summarize into JSON — Output a compact JSON data structure (see schema below)
9	3. Render & screenshot — Run `render.py` which injects JSON into `template.html` and screenshots to PNG
10
11	This design minimizes token cost: the HTML/CSS template is a static asset (~0 generation tokens). You only generate the JSON content.
12
13	## Step 1: Read the Article
14
15	- If given a URL: use `web_fetch` to retrieve content
16	- If given textual or image content: use it directly
17	- Read and deeply understand the article before summarizing
18
19	## Step 2: Save Poster Data as JSON
20
21	Write the poster data to `workspace/outputs/posters/poster_data.json` using the Write tool. Do NOT generate a Python script — just write the JSON file directly.
22
23	> IMPORTANT: Do NOT create a `gen_poster.py` or any intermediate HTML generation script. render.py handles template injection internally. Your only job is to write `poster_data.json`.
24
25	Follow this schema exactly:
26
27	```json
28	{
29	"source": "ANTHROPIC ENGINEERING",
30	"category": "HARNESS DESIGN",
31	"title": "Harness 设计：突破 Agent 编码前沿的关键架构",
32	"subtitle": "Anthropic 用一套受 GAN 启发的多智能体架构……核心不是更聪明的单 Agent，而是如何为强模型设计正确的 harness。",
33	"layout": "single",
34	"sections": [
35	{
36	"number": 1,
37	"color": "brown",
38	"title": "为什么简单方案会失效？",
39	"cards": [
40	{
41	"type": "text",
42	"heading": "上下文焦虑",
43	"body": "上下文一满，模型就会倾向于提前结束工作……"
44	},
45	{
46	"type": "text",
47	"heading": "自我评估偏差",
48	"body": "让模型评价自己的产出时……"
49	}
50	]
51	},
52	{
53	"number": 2,
54	"color": "olive",
55	"title": "架构核心：生成器 / 评估器分离",
56	"cards": [
57	{
58	"type": "highlight",
59	"heading": "受 GAN 启发，但不只是`多加一个 Agent`",
60	"body": "Anthropic 把`创造`与`评判`彻底解耦……"
61	}
62	]
63	},
64	{
65	"number": 4,
66	"color": "teal",
67	"title": "全栈开发 V1：三智能体分工",
68	"cards": [
69	{
70	"type": "tags",
71	"items": [
72	{ "tag": "Planner", "text": "把 1-4 句需求扩展成完整产品规格……" },
73	{ "tag": "Generator", "text": "按 Sprint 逐步实现……" },
74	{ "tag": "Evaluator", "text": "通过 Playwright 测运行中的应用……" }
75	]
76	}
77	]
78	},
79	{
80	"number": 5,
81	"color": "amber",
82	"title": "架构演进：从复杂到精简",
83	"cards": [
84	{
85	"type": "compare",
86	"left": { "heading": "V1：重型评估架构", "body": "单个 DAW 应用约 6 小时 / $200……" },
87	"right": { "heading": "V2：Claude 4.6 后减负", "body": "去掉 Sprint、把评估器后置……" }
88	}
89	]
90	},
91	{
92	"number": 6,
93	"color": "sage",
94	"title": "深层方法论",
95	"cards": [
96	{
97	"type": "bullets",
98	"items": [
99	"主观质量的可操作化：把`这设计好吗`转成`是否符合设计原则`。",
100	"结构化产物传递：重置上下文时，用文件而不是对话历史传状态。",
101	"持续压力测试：每个组件都编码了对模型能力的假设，而这些假设会随新模型迭代失效。"
102	]
103	},
104	{
105	"type": "callout",
106	"label": "关键洞察",
107	"body": "评估器是否必要，不取决于`它是不是好设计`，而取决于任务是否仍处在模型能力边界之外。"
108	}
109	]
110	}
111	]
112	}
113	```
114
115	### JSON Schema Rules
116
117	`layout` (optional) — Layout mode:
118	- `"single"`: Single column layout (default)
119	- `"double"`: Two column layout (odd sections left, even sections right)
120
121	`sections[]` — Any number of sections based on content needs. Each section has:
122	- `number`: Display number (1-based)
123	- `color`: One of `brown`, `olive`, `terracotta`, `teal`, `amber`, `sage`, `slate`, `rose`
124	- `title`: Section heading (concise, < 20 chars ideal)
125	- `cards[]`: Array of cards. Card types:
126
127	\| type \| fields \| description \|
128	\|------\|--------\|-------------\|
129	\| `text` \| `heading`, `body` \| Simple card with title + paragraph \|
130	\| `highlight` \| `heading`, `body` \| Card with colored left border accent \|
131	\| `tags` \| `items[{tag, text}]` \| Labeled tag + description list (like Planner/Generator/Evaluator) \|
132	\| `compare` \| `left{heading,body}`, `right{heading,body}` \| Side-by-side comparison boxes \|
133	\| `bullets` \| `items[]` \| Bullet list (supports bold: prefix) \|
134	\| `callout` \| `label`, `body` \| Highlighted callout box with label badge \|
135
136	### Content Guidelines
137
138	- Language: Match the user's language. Default Chinese for Chinese users.
139	- Conciseness: Each `body` should be 1-3 sentences. This is a poster, not a blog.
140	- Bold keywords: Use `keyword` for emphasis in body text.
141	- Section count: Use as many sections as needed to cover the content effectively (typically 3-10).
142	- Layout: Choose "single" for focused content or

People who liked this also viewed

6 picks

Type

article-poster

489stars57forks

View on GitHub

How to install article-poster?

magiccube/agentara/article-poster

$npx -y skills add MagicCube/agentara --skill article-poster

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Article Poster
2
3	Converts an article (URL or pasted text) into a beautifully designed infographic poster image.
4
5	## How It Works
6
7	1. Read the article — Fetch URL via `web_fetch` or use provided text
8	2. Summarize into JSON — Output a compact JSON data structure (see schema below)
9	3. Render & screenshot — Run `render.py` which injects JSON into `template.html` and screenshots to PNG
10
11	This design minimizes token cost: the HTML/CSS template is a static asset (~0 generation tokens). You only generate the JSON content.
12
13	## Step 1: Read the Article
14
15	- If given a URL: use `web_fetch` to retrieve content
16	- If given textual or image content: use it directly
17	- Read and deeply understand the article before summarizing
18
19	## Step 2: Save Poster Data as JSON
20
21	Write the poster data to `workspace/outputs/posters/poster_data.json` using the Write tool. Do NOT generate a Python script — just write the JSON file directly.
22
23	> IMPORTANT: Do NOT create a `gen_poster.py` or any intermediate HTML generation script. render.py handles template injection internally. Your only job is to write `poster_data.json`.
24
25	Follow this schema exactly:
26
27	```json
28	{
29	"source": "ANTHROPIC ENGINEERING",
30	"category": "HARNESS DESIGN",
31	"title": "Harness 设计：突破 Agent 编码前沿的关键架构",
32	"subtitle": "Anthropic 用一套受 GAN 启发的多智能体架构……核心不是更聪明的单 Agent，而是如何为强模型设计正确的 harness。",
33	"layout": "single",
34	"sections": [
35	{
36	"number": 1,
37	"color": "brown",
38	"title": "为什么简单方案会失效？",
39	"cards": [
40	{
41	"type": "text",
42	"heading": "上下文焦虑",
43	"body": "上下文一满，模型就会倾向于提前结束工作……"
44	},
45	{
46	"type": "text",
47	"heading": "自我评估偏差",
48	"body": "让模型评价自己的产出时……"
49	}
50	]
51	},
52	{
53	"number": 2,
54	"color": "olive",
55	"title": "架构核心：生成器 / 评估器分离",
56	"cards": [
57	{
58	"type": "highlight",
59	"heading": "受 GAN 启发，但不只是`多加一个 Agent`",
60	"body": "Anthropic 把`创造`与`评判`彻底解耦……"
61	}
62	]
63	},
64	{
65	"number": 4,
66	"color": "teal",
67	"title": "全栈开发 V1：三智能体分工",
68	"cards": [
69	{
70	"type": "tags",
71	"items": [
72	{ "tag": "Planner", "text": "把 1-4 句需求扩展成完整产品规格……" },
73	{ "tag": "Generator", "text": "按 Sprint 逐步实现……" },
74	{ "tag": "Evaluator", "text": "通过 Playwright 测运行中的应用……" }
75	]
76	}
77	]
78	},
79	{
80	"number": 5,
81	"color": "amber",
82	"title": "架构演进：从复杂到精简",
83	"cards": [
84	{
85	"type": "compare",
86	"left": { "heading": "V1：重型评估架构", "body": "单个 DAW 应用约 6 小时 / $200……" },
87	"right": { "heading": "V2：Claude 4.6 后减负", "body": "去掉 Sprint、把评估器后置……" }
88	}
89	]
90	},
91	{
92	"number": 6,
93	"color": "sage",
94	"title": "深层方法论",
95	"cards": [
96	{
97	"type": "bullets",
98	"items": [
99	"主观质量的可操作化：把`这设计好吗`转成`是否符合设计原则`。",
100	"结构化产物传递：重置上下文时，用文件而不是对话历史传状态。",
101	"持续压力测试：每个组件都编码了对模型能力的假设，而这些假设会随新模型迭代失效。"
102	]
103	},
104	{
105	"type": "callout",
106	"label": "关键洞察",
107	"body": "评估器是否必要，不取决于`它是不是好设计`，而取决于任务是否仍处在模型能力边界之外。"
108	}
109	]
110	}
111	]
112	}
113	```
114
115	### JSON Schema Rules
116
117	`layout` (optional) — Layout mode:
118	- `"single"`: Single column layout (default)
119	- `"double"`: Two column layout (odd sections left, even sections right)
120
121	`sections[]` — Any number of sections based on content needs. Each section has:
122	- `number`: Display number (1-based)
123	- `color`: One of `brown`, `olive`, `terracotta`, `teal`, `amber`, `sage`, `slate`, `rose`
124	- `title`: Section heading (concise, < 20 chars ideal)
125	- `cards[]`: Array of cards. Card types:
126
127	\| type \| fields \| description \|
128	\|------\|--------\|-------------\|
129	\| `text` \| `heading`, `body` \| Simple card with title + paragraph \|
130	\| `highlight` \| `heading`, `body` \| Card with colored left border accent \|
131	\| `tags` \| `items[{tag, text}]` \| Labeled tag + description list (like Planner/Generator/Evaluator) \|
132	\| `compare` \| `left{heading,body}`, `right{heading,body}` \| Side-by-side comparison boxes \|
133	\| `bullets` \| `items[]` \| Bullet list (supports bold: prefix) \|
134	\| `callout` \| `label`, `body` \| Highlighted callout box with label badge \|
135
136	### Content Guidelines
137
138	- Language: Match the user's language. Default Chinese for Chinese users.
139	- Conciseness: Each `body` should be 1-3 sentences. This is a poster, not a blog.
140	- Bold keywords: Use `keyword` for emphasis in body text.
141	- Section count: Use as many sections as needed to cover the content effectively (typically 3-10).
142	- Layout: Choose "single" for focused content or

People who liked this also viewed

6 picks

Type