用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/T0UGH/daily-report-master-agent --skill daily-report-lane-hacker-news命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | daily-report-lane-hacker-news |
| description | Generate the Hacker News section of the AI Agent daily report from raw corpus. |
HN 热门里和 AI/coding-agent 相关的帖子与评论讨论。
Read the lane package provided by the master:
input.mdcontext.jsonraw/history/ if presentDo not use selected_items.json as primary judgment input. Raw corpus is the evidence source.
Use recent reports only as reference-only dedupe context. Before selecting or writing, read yesterday and day-before-yesterday report files listed in context.json recent_report_paths or package history/. Reject exact repeats or substantially unchanged topics. Keep meaningful follow-ups with new facts and state what changed. Do not dedupe weather/current market items purely because yesterday had the same section. This is lane-agent judgment, not code-controlled filtering.
Select only items that can be explained concretely for an AI/coding-agent reader. Prefer specific releases, workflows, repos, discussions, failures, tools, versions, or user-visible changes.
Reject items when:
长度不是问题;问题是“有用信息密度”。宁可少选,也不要用低信息量内容填满栏目。每个入选条目必须同时满足:
写作时优先保留事实与判断依据;不要为了变短删除最有用的细节。
Include discussion substance, not title-only summaries. Explain what practitioners are debating, testing, doubting, or learning from comments/replies.
Write Chinese-first, concrete, human-readable prose. Keep source links.
子条目要信息密度高但不要拖长:
发生了什么 + 关键事实/数字 + 对读者有什么用/风险,少写铺垫和评价。禁止使用:
Write exactly:
lane.md: reader-facing markdown section only.lane-meta.json: JSON metadata with lane, status, selected_count, rejected_count, sources, rejected, and notes.
Allowed status: ok, empty, degraded, blocked.
Use selected and rejected reasoning in metadata so the master can audit the lane without rewriting it.