Tmtpost Summarizer

钛媒体（https://www.tmtpost.com/）文章阅读与资讯总结助手。支持四种模式：(1) 读取最新文章并深度总结；(2) 从快报频道汇总当前最新十条热点资讯；(3) 关键词过滤模式；(4) 指定URL直接总结；(5) 多篇横向对比分析。触发词：帮我看钛媒体、钛媒体快报、最新十条、关键词XX的钛媒体、...

Security Scan

The skill's code and instructions are consistent with its stated purpose (fetching and summarizing tmtpost.com articles); it fetches pages via curl, caches to a local ~/.cache directory, and does not request unrelated credentials or external services.

Purpose & Capability

Name/description (Tmtpost article summarizer) match the included scripts and SKILL.md. All network calls target tmtpost.com pages and the prompt templates describe article/nictation summarization. No unrelated services, credentials, or binaries are requested.

Instruction Scope

SKILL.md instructs the agent to fetch tmtpost URLs (homepage, /nictation, or user-provided article URLs) and then format summaries using the provided prompt template. The included shell scripts perform only site fetches, simple parsing, and local caching. They do not read unrelated system files or exfiltrate data to third-party endpoints.

Install Mechanism

No install spec or remote downloads; the skill is instruction-only with bundled shell scripts. Scripts are run locally (curl, grep, python3), which is expected for this functionality and poses normal local-execution risk but no hidden installer behavior.

Credentials

The skill requires no environment variables or credentials. It writes/reads a local cache directory (XDG_CACHE_DIR or $HOME/.cache/tmtpost), which is proportionate to its caching behavior. No secrets or unrelated config paths are requested.

Persistence & Privilege

always is false and the skill does not modify other skills or global agent settings. It stores cache files only under a per-user cache directory. No elevated or persistent privileges are requested.

Assessment

The skill appears coherent and limited to scraping and summarizing tmtpost.com pages. Before installing, review the bundled scripts (they will run curl, grep and python3) and note they will create/read a cache under XDG_CACHE_DIR or ~/.cache/tmtpost. The skill accepts user-provided URLs—avoid giving internal or sensitive URLs (SSRF / internal access risk if the agent environment can reach internal networks). Keywords are used as grep/regex patterns (may match more than expected) but there are no requested secrets or external endpoints. If you want extra safety, run the skill in a sandboxed environment or inspect/modify the scripts to restrict allowed hostnames to tmtpost.com.

钛媒体文章总结器

读取钛媒体最新文章或快报资讯，用专业科技媒体风格撰写结构化中文总结。支持四种模式。

四种模式

模式 A：最新文章深度总结（默认）

当用户说"帮我看钛媒体"、"总结 tmtpost"、"读一下这篇"时触发。

步骤：

1. web_fetch url="https://www.tmtpost.com/" extractMode="markdown" maxChars=5000
2. 从返回内容中找到最新文章 URL（时间最接近当前的文章）
3. web_fetch url="<文章URL>" extractMode="markdown" maxChars=20000
4. 按 references/prompt-template.md 单篇文章格式撰写总结

模式 B：最新十条快报汇总

当用户说"钛媒体快报"、"最新十条"、"快报"时触发。

步骤：

1. web_fetch url="https://www.tmtpost.com/nictation" extractMode="markdown" maxChars=15000
2. 可叠加 web_fetch url="https://www.tmtpost.com/fm" extractMode="markdown" maxChars=5000
3. 提取当前最新十条条目，按时间倒序
4. 按快报汇总格式输出

模式 C：关键词过滤快报

当用户说"关键词XX的钛媒体快报"、"只看我关注XX的钛媒体"时触发。

步骤：

1. 从用户 prompt 中提取关键词（可能有多个，用顿号、空格或逗号分隔）
2. 抓取 /nictation（如需更全可叠加 / 首页）
3. 在返回内容中搜索所有包含关键词的条目，精确匹配或语义相关均可
4. 如关键词为多条，先按相关度排序，再按时间倒序
5. 输出时在每条目前标注匹配的关键词

关键词支持示例：

• "AI、半导体的钛媒体快报"
• "人形机器人、新能源"
• "美团、字节"

模式 D：指定 URL 直接总结

当用户粘贴了具体文章 URL（https://www.tmtpost.com/数字.html）时触发。

步骤：

1. 直接 web_fetch url="<用户提供的URL>" extractMode="markdown" maxChars=20000
2. 按单篇文章格式输出总结

模式 E：多篇横向对比分析

当用户提供多个 URL、或说"钛媒体对比"、"横向对比"时触发。

步骤：

1. 抓取所有文章正文
2. 读取 references/prompt-template.md 中的对比分析格式
3. 输出一份结构化对比报告

脚本工具

# 最新文章 URL
./scripts/fetch-article.sh [max_articles]

# 快报（近十天）
./scripts/fetch-nictation.sh

# 搜索含关键词的快报（支持多词，逗号分隔）
./scripts/search-nictation.sh "AI,人形机器人"

错误处理

• 抓取失败 → 换用 /new 页面或重试
• 内容为空 → 不硬凑，换关键词或换数据源
• 关键词匹配为空 → 告知用户该关键词近期无内容，列出最接近的条目

触发词

模式	触发词
A 文章总结	"帮我看钛媒体"、"总结 tmtpost"、"钛媒体最新"
B 最新十条	"钛媒体快报"、"最新十条"、"快报"
C 关键词过滤	"AI的钛媒体快报"、"人形机器人关键词"
D 指定URL	直接粘贴文章链接
E 多篇对比	"钛媒体对比"、"横向对比分析"