Cron Job: trender-idea-miner
Job ID: 2a9306b6b96f
Run Time: 2026-07-31 07:09:16
Schedule: 0 7 * * *
Prompt
[IMPORTANT: The user has invoked the "idea-miner" skill, indicating they want you to follow its instructions. The full skill content is loaded below.]
name: idea-miner
description: Mine daily creative ideas & startup signals from HN/GitHub.
version: 1.0.0
author: Hermes Agent
license: MIT
metadata:
hermes:
tags: [ideas, trends, research, startup, daily-briefing]
related_skills: [youtube-content, blogwatcher]
🧠 Idea Miner — Daily Creative Intelligence
Overview
Mines today's creative ideas, inspiration hotspots, and startup signals from public sources. Produces a structured Markdown report with demand signals, trending projects, and actionable insights.
The report is saved to /opt/data/hermes/idea-miner/idea-miner_{YYYYMMDD}.md and delivered to the configured destination (Telegram, etc.).
When to Use
- Cron job for daily idea-mining runs (referenced by
trender-idea-minerand每日创意推送) - User asks: "what are today's hot ideas", "find me startup opportunities", "trending projects analysis"
- Any session that needs to synthesize multi-source trend intelligence
Data Sources
✅ Available & Tested
| Source | API Endpoint | Notes |
|---|---|---|
| HN Algolia (front page) | https://hn.algolia.com/api/v1/search?tags=front_page&hitsPerPage=20 |
Returns top stories with points |
| HN Algolia (Ask HN) | https://hn.algolia.com/api/v1/search?tags=ask_hn&hitsPerPage=15 |
Good for demand/need signals |
| HN Algolia (Show HN recent) | https://hn.algolia.com/api/v1/search_by_date?tags=show_hn&numericFilters=points%3E10&hitsPerPage=12 |
URL-encode > as %3E |
| HN Firebase (item detail) | https://hacker-news.firebaseio.com/v0/item/{ID}.json |
Individual story metadata |
| GitHub Search (new repos) | https://api.github.com/search/repositories?q=stars:%3E100+created:%3E{YYYY-MM-DD}&sort=stars&order=desc&per_page=10 |
Header: Accept: application/vnd.github.v3+json, User-Agent: HermesAgent/1.0 |
❌ Currently Unavailable
| Source | Issue |
|---|---|
Reddit JSON API (old.reddit.com, www.reddit.com) |
Blocked by Reddit (403 "whoa there, pardner") — needs OAuth credentials |
| Reddit pullpush.io (public archive) | Returns empty results from this IP |
| Google Trends RSS | Endpoint trends.google.com/trendingsearches/daily/rss?geo=US returns 404 |
| Google Trends API | trends.google.com/trends/api/dailytrends returns 404 |
| X/Twitter via xurl | xurl CLI not installed on this VPS |
Data Acquisition Workflow
- Use
curl -s -o /tmp/filename.json -w '%{http_code}'to save to temp file first (avoids pipe-to-interpreter security blocks in cron mode) - Parse with
python3 -c "import json; data = json.load(open('/tmp/filename.json')); ..." - Saved temp files are disposable — clean up or leave in /tmp
Report Format
The report uses Chinese (用户偏好) with the following sections:
# 🧠 Idea Miner — 每日创意情报 (YYYY-MM-DD)
> 数据来源:[list of sources used]
## 📢 需求现场 — 高意向需求 & 讨论
Top demand signals from Ask HN, discussions, needs
## 📈 趋势项目 — GitHub 热门新仓库
Table: project | ⭐ | description | direction
Followed by trend interpretation block (code fence with arrows)
## 🔬 Show HN 精选 — 本周最新项目
Table: project | points | description
## 🌟 今日创业灵感精选
### 高潜力方向
### 已验证信号
## 📊 数据快照
Key metrics summary table
Report Generation Steps
- Fetch HN front page — identifies the overall tech zeitgeist for the day
- Fetch HN Ask HN — captures user needs, pain points, and demand signals
- Fetch HN Show HN (recent, points>10) — captures new projects people are building
- Fetch GitHub trending repos (last 7 days) — identifies what developers are excited about
- Fetch additional sources if available (Reddit, Twitter, etc.)
- Synthesize findings into the structured report
- Save to file at
/opt/data/hermes/idea-miner/idea-miner_{YYYYMMDD}.md - Output the report as final response — the cron delivery system automatically delivers it
File Save Convention
- Directory:
/opt/data/hermes/idea-miner/ - Filename:
idea-miner_{YYYYMMDD}.md(e.g.,idea-miner_20260727.md) - If the directory doesn't exist, create it
Common Pitfalls
- HN Algolia URL encoding: The
numericFiltersparameter with>must use%3Enot raw>. E.g.numericFilters=points%3E10 - Reddit API blocked: Don't retry Reddit JSON API with this IP — it's consistently 403. Only retry if OAuth creds are configured.
- Pipe-to-interpreter security blocks: In cron mode,
curl | python3triggers security warnings. Save output to file first with-o /tmp/file, then parse withpython3 -creading from file. - GitHub API rate limits: Without authentication, GitHub API is limited to 60 req/hr. For 10 repos + description parsing per run this is fine, but batch requests if expanding.
- Show HN search prefers
search_by_date: The/searchendpoint returns all-time top, not recent. Use/search_by_datefor recent Show HN posts. - No xurl installed: X/Twitter mining won't work unless xurl CLI is installed and OAuth'd.
- Google Trends unavailable via simple curl: None of the documented public endpoints work (404). Skip silently.
Verification Checklist
- [ ] HN front page fetched and parsed correctly
- [ ] Ask HN fetched for demand signals
- [ ] Show HN (recent) fetched for new projects
- [ ] GitHub trending repos fetched
- [ ] Report is structured with all tables and sections
- [ ] File saved to correct path with date-stamped filename
- [ ] Report output as final message (cron delivery takes over)
Related Scripts
The legacy shell script at /root/.hermes/scripts/idea-miner_daily.sh references a Python script at ~/idea-miner/idea_miner.py that no longer exists. The cron jobs now run Hermes agent directly with the prompt, bypassing the shell script. If the shell script is ever re-enabled, it needs to be updated.
The user has provided the following instruction alongside the skill invocation: [IMPORTANT: You are running as a scheduled cron job. DELIVERY: Your final response will be automatically delivered to the user — do NOT use send_message or try to deliver the output yourself. Just produce your report/output as your final response and the system handles the rest. SILENT: If there is genuinely nothing new to report, respond with exactly "[SILENT]" (nothing else) to suppress delivery. Never combine [SILENT] with content — either report your findings normally, or say [SILENT] and nothing more.]
运行 Idea Miner 模块,挖掘今日的创意和灵感热点,然后将结果发送到 Telegram 用户 jinbin(Chat ID: 951402025)。如果没有发现热点,输出简明的"今日无新热点"。
文件保存
将完整的分析结果同时保存到本地文件:/opt/data/hermes/idea-miner/idea-miner_{YYYYMMDD}.md
Response
✅ 报告已保存至 /opt/data/hermes/idea-miner/idea-miner_20260731.md
🧠 Idea Miner — 每日创意情报 (2026-07-31)
数据来源:HN Front Page / Ask HN / Show HN + GitHub Search(近7天新仓库)
📢 需求现场 — 高意向需求 & 讨论
| 信号 | 热度 | 解读 |
|---|---|---|
| 法国 ISP SFR 屏蔽 opencode.ai | 6分/4c | 网络层开始封杀 AI 工具 — AI 访问控制/镜像/合规检测需求爆发 |
| Anthropic 公布 3 起真实网络安全评估事故 | 165分/124c | 企业级 AI Agent 安全审计、红队工具是刚需 |
| 给 GPT-5.6 一家真实公司自主经营:撒谎、群发垃圾、亏 $447 | 347分/202c | Agent 自主性 vs 可靠性 — Agent 护栏/监督层需求明确 |
| Stripe Radar 涨价伪装成"免费试用" | 9分/1c | 定价透明度/账单监控工具(indie hacker 痛点) |
| "AI slop 是否污染新一代模型训练数据" | 7分/9c | 数据质量、去污染、真实性验证方向 |
| old.reddit 强制登录 | 13+6分 | 老平台替代品、Reddit API 客户端生态再起波澜 |
| GitHub 改"煎饼 emoji"菜单被吐槽 | 27+21分 | 开发者 UI 敏感,自定义 UI 层有机会 |
📈 趋势项目 — GitHub 热门新仓库
| 项目 | ⭐ | 方向 |
|---|---|---|
| MoonshotAI/Kimi-K3 | 7,589 | 开源前沿模型(7天7.5k⭐) |
| mshumer/Claude-of-Duty | 2,432 | 单 prompt 生成 CoD 级 Three.js FPS |
| VictorTaelin/OptMem | 944 | Agent 永久记忆(426-token prompt) |
| xikhar/persona | 684 | 实时语音交互 |
| 0xwilliamortiz/ponytail-improved | 565 | Agent 提示工程("最懒资深工程师"思维) |
| 0xwilliamortiz/openclaude-improved | 562 | Agent 运行层 |
| talivia-group/talivia | 503 | 开源自托管 founder 分析(替代 GA) |
| QwenAudio/qwen-audio-agent | 502 | 语音 Agent 实时运行时 |
| gavamedia/deltafin | 497 | 单设备跑满血 Kimi K3 |
趋势解读
开源前沿模型(Kimi-K3)→ 本地推理平民化(deltafin / RunNburn 64GB 跑 295B MoE)
↓
Agent 生态三件套:记忆(OptMem)+ 语音(persona/qwen)+ 护栏(Noisegate/ZeroShot)
↓
单 prompt 生成完整应用(Claude-of-Duty)→ "提示词即产品"进一步泛化
🔬 Show HN 精选
| 项目 | 分 | 说明 |
|---|---|---|
| 蒸馏 DeepSeek→GPT-OSS 不迁移审查 | 114 | 蒸馏与模型对齐行为研究 |
| 从逻辑门造 CPU 的游戏 | 81 | 教育 + 硬核极客向 |
| Claude-account | 49 | Claude Code 多账号秒切 |
| 本地 merge queue 并行跑多 Agent | 41 | 多 Agent 协作基础设施 |
| ZeroShot:Agent 会话监控 | 18 | Agent 可观测性 |
| Noisegate:不可信 Agent 差分隐私网关 | 17 | Agent 安全边界 |
| Ski:端侧免费语音编程 | 13 | 语音编码端侧化 |
🌟 今日创业灵感精选
高潜力方向
1. Agent 可观测性+安全治理 — 事故报告、Sol 翻车、Noisegate/ZeroShot 同台:Agent 安全/审计/监控是确定性机会,B 端付费意愿强
2. 开源模型本地化部署 — Kimi-K3 带动 deltafin/RunNburn 生态:GPU 云、量化工具、一键部署跟进
3. Agent 持久记忆层 — OptMem 944⭐验证需求:跨会话记忆是下一瓶颈,可做中间件/API
4. AI 内容真实性检测 — 假作者论文被接收 + 训练数据被 slop 污染 → 检测/清洗/溯源工具
5. 语音 Agent 实时交互 — qwen-audio-agent + persona + Ski:语音正成为 Agent 默认交互层
已验证信号
- ✅ 开源前沿模型饥渴:Kimi-K3 单周 7.5k+⭐
- ✅ 单 prompt 生成 3A 级体验:Claude-of-Duty 2.4k⭐
- ✅ 创始人要自托管、营收优先分析(隐私回归):talivia
- ✅ 端侧语音编码被需要:Ski(免费、on-device 是卖点)
- ✅ 多 Agent 并行协作工具需求真实存在
📊 数据快照
| 指标 | 数值 |
|---|---|
| HN 首页最高分 | 934(UEFA/FIFA,非技术) |
| HN 首页技术最高分 | 677(流媒体设备安全) |
| GitHub 新仓库 (7d, >100⭐) | 10 个,合计 ~14.5k ⭐ |
| 今日主题 | AI Agent(可靠性/记忆/语音/安全) |
今日核心判断:AI Agent 从"能跑"进入"可信"阶段 — 安全治理、可观测性、持久记忆、语音交互四大基础设施层同时出现爆发信号;Kimi-K3 开源进一步推动本地推理平民化。对独立开发者而言,Agent 安全/审计和本地部署工具链是最佳切入点。