メニュー
Use when generating a daily 24h AI / Agent / AIGC top-news briefing, publishing it as a Feishu/Lark native document, validating it, and optionally archiving it to Bitable.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/dracohu2025-cloud/draco-skills-collection --skill daily-ai-agent-aigc-top-newsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Use when generating PPT-style image slides, poetic presentation covers, quiet paper-texture visual pages, report pages, invitations, social cards, or slide-image sets with GPT-Image-2 via image_generate.
Use when generating high-readability hand-drawn knowledge diagrams, architecture diagrams, workflow maps, or consulting-style visual explanations with GPT-Image-2 via image_generate.
可独立运行的 GPT-Image 增强版 EPUB2Podcast：在本地把 EPUB 转成双人中文音频、GPT-Image/Smart Slide 视觉页、最终 MP4，并生成 YouTube 发布素材。
SOC 職業分類に基づく
SKILL.md を表示中
| name | daily-ai-agent-aigc-top-news |
| description | Use when generating a daily 24h AI / Agent / AIGC top-news briefing, publishing it as a Feishu/Lark native document, validating it, and optionally archiving it to Bitable. |
| version | 1.0.0 |
| author | Hermes Agent |
| license | MIT |
| platforms | ["linux"] |
| prerequisites | {"commands":["python3","lark-cli"]} |
| metadata | {"hermes":{"tags":["news","ai","agent","aigc","briefing","cron","feishu","lark","bitable"],"related_skills":["news-aggregator-skill","feishu-lark-workflows","ai-news-bitable-archive"]}} |
This skill runs a daily AI / Agent / AIGC morning briefing workflow.
It collects candidates from news aggregators and source-of-truth pages, filters for high-signal items, writes a Chinese 24h report, publishes it as a Feishu/Lark native document, fetches it back for validation, and optionally archives the result to Feishu Bitable.
Core rule: truth first, no quota-filling. If there are only a few important items, keep the report short.
Use this skill when the task asks for any of these:
Do not use this for:
A cron prompt should be self-contained and include:
Task: Generate the past-24h AI / Agent / AIGC Top News briefing.
Timezone: Asia/Shanghai.
Delivery: create Feishu/Lark native doc, fetch-back validate, optionally archive to Feishu Bitable, send result to origin chat.
Archive base_token: <FEISHU_BITABLE_BASE_TOKEN>
Archive table_id: <FEISHU_BITABLE_TABLE_ID>
Bitable URL: https://www.feishu.cn/base/<FEISHU_BITABLE_BASE_TOKEN>?table=<FEISHU_BITABLE_TABLE_ID>
Required project check: <owner>/<repo>
If no Feishu folder token is supplied, create the document in the default Feishu location. Do not block.
Final answer must include:
已完成。
- 文档标题：...
- doc_url：...
- bitable_url：... # if archival is enabled
- record_id：... # if archival is enabled
- 摘要：
- ...
- ...
- ...
If archival fails but doc publishing succeeds, say so explicitly and include the doc URL plus a short error summary.
Always check live time and Feishu/Lark auth. Do not infer time mentally.
date -u '+UTC=%Y-%m-%d %H:%M:%S' && TZ=
lark-cli auth status
lark-cli auth scopes
Minimum expected capabilities:
If scope output is noisy, continue if the actual create/fetch/API calls work.
Prefer a local news aggregator as the first pass. It gives candidates, not final truth.
Example commands, adapt paths to your local setup:
python3 path/to/news_aggregator_run.py smoke-test --quick
python3 path/to/news_aggregator_run.py fetch --source hackernews --limit 30 --deep --save --outdir /tmp/<run>/hn
python3 path/to/news_aggregator_run.py fetch --source github --limit 30 --deep --save --outdir /tmp/<run>/github
python3 path/to/news_aggregator_run.py fetch --source huggingface --limit 30 --deep --save --outdir /tmp/<run>/hf
Also perform targeted checks against official/source-of-truth pages. At minimum check:
Use web search/extract for current facts. If a page blocks extraction, combine search-result snippets, official docs/API pages, and one reputable secondary source.
Always check the configured project separately. External popularity does not decide whether it appears.
Default recommended project for this workflow:
NousResearch/hermes-agent
REPO_DIR="/path/to/local/repo"
if git -C "$REPO_DIR" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
git -C "$REPO_DIR" fetch origin --prune
git -C "$REPO_DIR" log origin/main --since='24 hours ago' --pretty=format:'%h | %ad | %s' --date=iso
fi
python3 - <<'PY'
import json, urllib.request
owner_repo = 'NousResearch/hermes-agent'
req = urllib.request.Request(
f'https://api.github.com/repos/{owner_repo}/releases/latest',
headers={'User-Agent':'Daily-AI-Agent-Briefing'}
)
with urllib.request.urlopen(req, timeout=30) as r:
data = json.load(r)
print(json.dumps({k:data.get(k) for k in ['tag_name','name','published_at','html_url']}, ensure_ascii=False, indent=2))
PY
python3 - <<'PY'
import json, urllib.parse, urllib.request, datetime
owner_repo = 'NousResearch/hermes-agent'
start = (datetime.datetime.now(datetime.timezone.utc)-datetime.timedelta(hours=24)).date().isoformat()
q = f'repo:{owner_repo} is:pr merged:>={start} sort:updated-desc'
url = 'https://api.github.com/search/issues?' + urllib.parse.urlencode({'q':q,'per_page':10})
req = urllib.request.Request(url, headers={'User-Agent':'Daily-AI-Agent-Briefing'})
with urllib.request.urlopen(req, timeout=30) as r:
data = json.load(r)
for item in data.get('items',[])[:10]:
print(json.dumps({'number':item['number'],'title':item['title'],'html_url':item['html_url']}, ensure_ascii=False))
PY
Include the checked project if any of these occurred in the window:
If there is no material update, still mention that the check was performed in source notes or omit the section only if truly empty. Never skip the check.
Use this priority order:
Hard rules:
objectID / hn_url; do not guess item IDs.Title format:
AI / Agent / AIGC Top News 24h｜YYYY-MM-DD 08:00
Markdown body template:
更新于：YYYY-MM-DD 08:00 CST
统计窗口：过去24小时
筛选口径：官方发布 / GitHub release 或 PR / 权威媒体 / 开源趋势信号。Trending 只作为趋势，不等同于正式发布。
最终保留：N 条。没硬凑。
## 最值得注意的 3 条
### 1. 标题
发生了什么：...
为什么重要：...
对工作流影响：...
来源：...
### 2. 标题
...
### 3. 标题
...
## 模型 / Agent 产品
## AIGC 生图 / 生视频
## 评测 / 基准 / 研究
## GitHub Trending / 开源趋势信号
## 开源项目 / Toolchain 信号
## 指定项目过去24小时更新
## 产业动态
## 一句话结论
...
Formatting rules:
## heading must have a blank line before it.Prefer writing the report to a local temporary Markdown file, then pass the file content via Python/subprocess or safe argv. Avoid shell-embedded huge markdown.
Minimal create command:
lark-cli docs +create \
--as user \
--title 'AI / Agent / AIGC Top News 24h｜YYYY-MM-DD 08:00' \
--markdown '<content>'
If a folder token is provided:
lark-cli docs +create \
--as user \
--folder-token '<FEISHU_FOLDER_TOKEN>' \
--title 'AI / Agent / AIGC Top News 24h｜YYYY-MM-DD 08:00' \
--markdown '<content>'
After creation, extract doc_url from stdout.
Always fetch the created document before reporting success.
lark-cli docs +fetch --as user --doc '<doc_url>' --format pretty
Validate:
## 最值得注意的 3 条 exists## AIGC 生图 / 生视频 exists## GitHub Trending / 开源趋势信号 exists## 一句话结论 existsPretty fetch may normalize ordered lists to 1.. That is display-layer behavior; do not treat it as failure if structure is intact.
Use a dedicated archive script if available. Do not hand-build shell JSON.
Example command:
python3 path/to/sync_doc_to_bitable.py \
--doc-url '<doc_url>' \
--base-token '<FEISHU_BITABLE_BASE_TOKEN>' \
--table-id '<FEISHU_BITABLE_TABLE_ID>' \
--date 'YYYY-MM-DD' \
--status '已归档'
Bitable URL format:
https://www.feishu.cn/base/<FEISHU_BITABLE_BASE_TOKEN>?table=<FEISHU_BITABLE_TABLE_ID>
The script should return record_id. Fetch that exact record:
lark-cli api GET "/open-apis/bitable/v1/apps/<FEISHU_BITABLE_BASE_TOKEN>/tables/<FEISHU_BITABLE_TABLE_ID>/records/<record_id>" --as user
Validate fields:
If the table schema changes, inspect fields:
lark-cli api GET "/open-apis/bitable/v1/apps/<FEISHU_BITABLE_BASE_TOKEN>/tables/<FEISHU_BITABLE_TABLE_ID>/fields?page_size=100" --as user
Remember: Feishu URL fields require {link, text}, not a raw string.
任务：生成“过去24小时 AI / Agent / AIGC Top News 早报”，发布到飞书原生文档，并在发布成功后自动归档到飞书多维表。完成后把结果回传到当前聊天。
必须使用已加载的 daily-ai-agent-aigc-top-news skill 执行完整流程。
固定要求：
1. 时间窗口：过去24小时，时区 Asia/Shanghai。
2. 内容范围：AI / Agent / coding agent / eval / workflow / toolchain / AIGC 生图生视频。
3. 必须单独核查 <owner>/<repo>：release、最近24小时 commits、merged PR、重要用户可感知变化。
4. 必须检查 GitHub Trending Today；只能写成“今日趋势信号”，不能冒充过去24小时正式发布。
5. 必须检查 AIGC 生图 / 生视频官方源：OpenAI、Google、Runway、Pika、Kling、字节/即梦/Seedance、Midjourney、Ideogram、Adobe Firefly、Stability。
6. 真实性优先；不要硬凑条数。
7. 输出中文。
8. 发布为飞书原生文档，标题：AI / Agent / AIGC Top News 24h｜YYYY-MM-DD 08:00。
9. 创建后必须 docs +fetch 回读验收。
10. 如启用归档，归档到飞书多维表：base_token <FEISHU_BITABLE_BASE_TOKEN>，table_id <FEISHU_BITABLE_TABLE_ID>。
11. 如启用归档，归档后必须按 record_id 回读验收。
12. 最终回复包含：文档标题、doc_url、bitable_url、record_id、3～6条摘要。
Attach these skills to the cron job if available:
daily-ai-agent-aigc-top-news
news-aggregator-skill
ai-news-bitable-archive
feishu-lark-workflows
Only using the aggregator. Aggregator gives candidates; official/GitHub/API evidence decides inclusion.
Skipping AIGC media checks. Daily report must cover image/video model and product updates, even when the final answer says no material update.
Treating Trending as release news. GitHub Trending is a trend signal only.
Guessing HN item IDs. Always use objectID / hn_url from source data.
Forgetting the mandatory project check. It is a separate check, not just another candidate.
Shell JSON pain. For Feishu/Bitable, prefer Python subprocess with argv and JSON serialization.
Reporting before validation. Doc creation alone is not enough. Fetch back the doc and the Bitable record.
Writing glued Markdown. Keep blank lines between intro paragraphs and before every ## heading.
docs +fetch validation passedrecord_id, if archival is enabled