name: ai-wiki-cook-blog
description: 用户给出单篇公开可见 blog/article URL，并要求 cook、烹饪或消化成 AI wiki human inbox 笔记和必选信息图时使用；浏览器优先，同 URL 公开 HTML/readability fallback，不进入 canonical wiki ingest。
from skill_contract import *
skill(
name="ai-wiki-cook-blog",
purpose="将单篇公开可见博客文章烹饪成带信息图的 AI wiki human inbox 笔记。",
)
activate_when([
"用户给出单篇公开可见 blog、博客文章、article 或网页长文 URL，并要求 cook、烹饪、消化或整理成 AI wiki human inbox 笔记",
])
do_not_activate_when([
"用户给出 X/Twitter、GitHub repository、Apple Podcasts 或 cook my mind 输入；改用对应 cook skill",
"用户要求 ingest、compile、source coverage、human/raw 处理、source note、index 或 log 更新",
"用户要求抓取整个网站、从首页发现文章、RSS/feed 解析、系列综述、外链研究或事实核验",
"输入不是单篇文章 URL，而是主页、目录页、搜索结果、tag 页、newsletter archive 或聚合页",
])
inputs(
required=[
input("article_url", type=URL, description="要 cook 的单篇公开博客文章 URL。"),
],
optional=[
input("user_focus", type=NaturalLanguage, description="用户特别希望关注的问题、角度或用途。", required=False),
input("title_preference", type=Text, description="用户给出的笔记标题偏好。", required=False),
],
)
outputs(
required=[
output("capture_record", type=File, description=".codex/cache/cook-blog/<url-hash>/capture.md，结构化观察和限制记录，不保存完整正文。"),
output(
"cooked_note",
type=File,
description="human/inbox/cook-blog/YYYY-MM-DD_<中文主题标题>_<站点或作者>.md。",
required_sections=[
"速读",
"原文",
"内容地图",
"关键论点",
"核心内容",
"关键洞察",
"批判性点评",
"对我的启发",
"可以继续追的问题",
"信息图",
"遗漏与不确定",
"Source Manifest",
],
),
output("infographic_asset", type=File, description="human/inbox/cook-blog/assets/<note-stem>/infographic.webp。"),
output("source_manifest", type=Text, description="最终 note 中的 Source Manifest。"),
],
)
environment(
commands=["python3", "rg"],
dependencies=["browser automation", "public HTML/readability extraction when needed", "image generation"],
network="required",
filesystem="workspace",
)
decision_rules([
when("article_url is missing", then="ask for exactly one public article URL"),
when("input is a neighboring cook type", then="stop and route to the matching cook skill"),
when("input is not a single article URL", then="block and ask for a single article URL; do not discover articles automatically"),
when("browser capture exposes enough main article body", then="use browser capture as source of truth", else_="try same-URL public HTML/readability fallback"),
when("fallback is needed", then="fetch only the input URL or a page-declared canonical same-article URL; do not search, open mirrors, use third-party caches, or open external links"),
when("browser capture and fallback both fail", then="write optional blocker capture.md and stop without infographic or final note"),
when("imagegen, cache-original persistence, or final infographic.webp persistence fails", then="block and do not write the final note"),
])
workflow([
step(
"read_project_rules",
"Read AGENTS.md human boundary, human/inbox workflow, and non-canonical cook output rules before writing any output.",
writes=["project_rules"],
),
step(
"classify_and_prepare",
"Confirm article_url is one public article, reject neighboring cook and non-article inputs, then derive .codex/cache/cook-blog/<url-hash>/ and human/inbox/cook-blog/ paths.",
reads=["article_url", "title_preference"],
writes=["input_classification", "cache_paths", "note_paths"],
),
step(
"capture_article",
f"""
Capture only the main article surface: title, site, author, date, headings, body structure, short evidence excerpts, body links, images, code, and tables. Exclude comments, recommendations, ads, signup forms, navigation, footer, and unrelated page chrome.
Use {call_tool(
"browser automation",
how="open article_url without login or account state, wait for the page to settle, inspect the public visible article, and judge whether the main body is readable",
expect="structured browser observations or a clear browser blocker",
on_failure="record the browser failure and try same-URL public HTML/readability fallback",
)}.
If browser observations are insufficient, use {call_tool(
"python3 public HTML/readability fallback",
how="fetch only article_url or a visibly declared canonical same-article URL; extract structured article observations without saving full extracted text",
expect="structured article observations or a clear blocker",
on_failure="write optional blocker capture.md and stop without final note or infographic",
)}.
""",
reads=["article_url", "input_classification"],
writes=["article_observations"],
),
step(
"write_capture_record",
"Write capture.md with input URL, canonical URL if known, captured_at, capture method, browser/readability actions, article outline, key claims, short excerpts, body links marked not verified, media/code/table observations, exclusions, limitations, and blockers. Do not preserve full article text.",
reads=["article_observations", "cache_paths"],
writes=["capture_record"],
),
step(
"stop_on_blocker",
"If capture_record shows unreadable, paywalled, login-gated, non-article, or too-sparse content, report the blocker and cache evidence path; do not generate an infographic or final note.",
reads=["capture_record"],
),
step(
"draft_note",
"Write the Chinese cooked note draft from capture_record. Preserve the original URL, include a content map, distinguish 作者明确说法 / Agent 推断 / 我的启发, include critical commentary, and apply user_focus only when supported by the article evidence.",
reads=["capture_record", "user_focus", "project_rules"],
writes=["cooked_note_draft"],
),
step(
"generate_and_persist_infographic",
f"""
Generate and persist the required Chinese infographic.
Use {call_tool(
"imagegen",
how="create a bitmap infographic from the cooked understanding; make it a review aid covering core thesis, structure, claims, tensions, and useful implications",
expect="imagegen original copied to cache and final infographic.webp written to human/inbox/cook-blog/assets/<note-stem>/",
on_failure="block and do not write the final note",
)}.
""",
reads=["cooked_note_draft", "cache_paths", "note_paths"],
writes=["infographic_asset"],
),
step(
"write_final_note",
"Write the final Obsidian Markdown note under human/inbox/cook-blog/ with cook-blog frontmatter, required sections, infographic wikilink, limitations, and Source Manifest. Do not update canonical graph files unless the user separately asks for ingest or compile.",
reads=["cooked_note_draft", "capture_record", "infographic_asset", "note_paths"],
writes=["cooked_note", "source_manifest"],
),
step(
"report_result",
"Report note path, infographic path, cache path, and limitations; if blocked, report blocker and cache evidence path instead.",
reads=["cooked_note", "infographic_asset", "capture_record"],
),
])
quality_bar(
must=[
"只处理单篇公开可见博客文章 URL。",
"只写 human/inbox/cook-blog/ 和 .codex/cache/cook-blog/；不进入 canonical ingest。",
"浏览器优先；fallback 只能读取同 URL 或页面声明的同文章 canonical URL。",
"不登录、不用账号态、不搜索、不换来源、不打开外链、不用第三方缓存或镜像。",
"只消费主文章正文；排除评论、推荐、广告、signup、导航、页脚和无关页面外壳。",
"capture.md 只保存结构化观察、关键论点、少量短摘录、正文链接线索和限制，不保存完整正文。",
"最终 note 为中文，frontmatter 使用 type: cook-blog、ingest_policy: on-request、inbox_status: unread、source_kind: blog-article、source_url、captured_at。",
"最终 note 包含 required_sections，并在关键论点中区分作者明确说法、Agent 推断和我的启发。",
"外链只记录可见 anchor text 和 URL，并标注未核验。",
"信息图必选；imagegen、cache 原图或 infographic.webp 任一步失败，都不写最终 note。",
"Source Manifest 列出 input URL、canonical URL if known、capture method/actions、cache paths、infographic path、exclusions and limitations。",
],
should=[
"内容地图优先使用 Mermaid mindmap；流程、因果链或决策树可用 flowchart。",
"文件名优先中文可读，末尾保留站点名、作者或关键英文术语。",
"作者、日期或 canonical URL 不可见时继续 cook，并在遗漏与不确定中说明。",
],
must_not=[
"不要更新 index.md、log.md、sources/、human/sources/、concepts/、entities/、synthesis/、maps 或 questions，除非用户明确另行要求 ingest 或 compile。",
"不要处理站点首页、系列页、搜索结果、tag 页、RSS/feed、newsletter archive 或多个 URL。",
"不要把评论、推荐阅读或作者 bio 当成主文章论据。",
"不要把 Agent 推断写成作者原话。",
"不要声称外链或外部事实已经核验。",
],
)
By