Merge OrbitOS into project — one install, everything ready

- Added 15 OrbitOS skills (bundled in skills/orbitos/) - Added vault structure templates (views, templates, prompts) - Added install.sh (macOS + Linux) - Removed references/orbitos-skills.md (no longer needed) - Updated README: 30 skills total, unified install flow Total: 9 mm* + 6 dependencies + 15 orbitos = 30 skills
2026-07-02 10:13:46 +07:00
parent 7bf147069e
commit ddac3f187c
45 changed files with 2695 additions and 197 deletions
--- a/skills/orbitos/orbites-ai-newsletters/SKILL.md
+++ b/skills/orbitos/orbites-ai-newsletters/SKILL.md
@@ -0,0 +1,209 @@
+---
+name: orbites-ai-newsletters
+description: Curate daily AI, Online Marketing, SEO, and AI×Marketing Research news digest. For OrbitOS. Use when user says 'AI news', 'marketing news', 'SEO news', 'newsletter digest', 'what's new in AI', 'ข่าว AI', 'ข่าว marketing', 'สรุปข่าว'.
+---
+# OrbitOS: Daily News Digest — AI · Marketing · SEO · Research
+
+Fetch, deduplicate, and rank news from 5 curated topic areas into a daily digest saved to the OrbitOS vault.
+
+## Source Configuration
+
+### 🔵 AI News
+| Source | URL | Type |
+|--------|-----|------|
+| TLDR AI | `https://bullrich.dev/tldr-rss/ai.rss` | RSS |
+| The Rundown AI | `https://rss.beehiiv.com/feeds/2R3C6Bt5wj.xml` | RSS |
+
+### 🟢 Online Marketing
+| Source | URL | Type |
+|--------|-----|------|
+| Marketing AI Institute | `https://www.marketingaiinstitute.com/blog/rss.xml` | RSS |
+| HubSpot Marketing | `https://blog.hubspot.com/marketing/rss.xml` | RSS |
+| Search Engine Land | `https://searchengineland.com/feed` | RSS |
+
+### 🟡 SEO
+| Source | URL | Type |
+|--------|-----|------|
+| Search Engine Journal | `https://www.searchenginejournal.com/feed/` | RSS |
+| Ahrefs Blog | `https://ahrefs.com/blog/feed/` | RSS |
+| Semrush Blog | `https://www.semrush.com/blog/feed/` | RSS |
+| Moz Blog | `https://feedpress.me/mozblog` | RSS |
+| Neil Patel | `https://neilpatel.com/blog/feed/` | RSS |
+
+### 🟣 Marketing & AI Research
+| Source | URL | Type |
+|--------|-----|------|
+| arXiv (AI + marketing) | `https://export.arxiv.org/api/query?search_query=all:marketing+AND+all:AI&sortBy=submittedDate&sortOrder=descending&max_results=10` | API |
+| Web search | Execute: `web_search("AI marketing research 2026 generative AI")` + `web_search("marketing AI agent study 2026")` | Search |
+
+## Workflow
+
+1. **Check cache**: Look for `50_Resources/Newsletters/YYYY-MM/YYYY-MM-DD-Digest.md` (relative to vault root). If exists with today's date, return cached content.
+
+2. **Fetch feeds** (see Provider Fallbacks for exact curl commands):
+   - RSS sources: Use `curl` via terminal on each URL. Parse title, link, pubDate, description from XML. Fall back to `web_extract` if configured.
+   - arXiv: Use `curl` on the API URL. Parse `<entry>` elements.
+   - Web search: Run via `web_search` if configured; otherwise skip this section.
+
+3. **Classify each item** into topic buckets:
+   - `ai` — LLMs, GPT, Claude, agents, ML, AI tools
+   - `marketing` — digital marketing, paid ads, content strategy, brand, B2B marketing
+   - `seo` — search rankings, Google updates, SERP features, technical SEO, GEO
+   - `research` — academic papers, studies, surveys, reports with data/statistics
+
+4. **Deduplicate**: Items with 70%+ title word overlap across sources → merge into one entry tracking all source URLs.
+
+5. **Rank within each bucket** by:
+   - Relevance to the topic cluster
+   - Content creation potential (can I write a post about this?)
+   - Novelty (penalize if similar item in recent archives)
+   - Recency
+
+6. **Generate digest** — bilingual Thai/English:
+
+```markdown
+# 📰 Daily Digest — YYYY-MM-DD
+
+## 🔵 AI — Top 3
+- **[Title]** — 1-line summary. [Source]
+  → *Angle:* content idea in Thai
+...
+
+## 🟢 Online Marketing — Top 3
+...
+
+## 🟡 SEO — Top 3
+...
+
+## 🟣 Marketing & AI Research — Top 2
+...
+
+## 💡 Content Creation Opportunities
+Top 5 across all buckets ranked by post potential:
+1. ...
+```
+
+7. **Save files** (all paths relative to vault root):
+   - `50_Resources/Newsletters/YYYY-MM/YYYY-MM-DD-Digest.md` — curated digest
+   - `50_Resources/Newsletters/YYYY-MM/Raw/` — raw extracted content per source if substantial
+
+8. **Curate high-value articles → migrate to 30_Research/**
+   - After generating the digest, scan Content Creation Opportunities and related articles
+   - For each article worth keeping for future content creation:
+     1. Copy/save as `30_Research/<slug-title>.md` with frontmatter `type: article` + source URL
+     2. Remove from `Raw/` folder (don't keep raw dumps there long-term)
+     3. Add a note in the Research file linking back to the digest
+   - The user confirmed this: **`Raw/` = processing cache, `30_Research/` = permanent collection**
+   - Pitfall: Don't let Raw/ accumulate — it's a temp folder, not an archive
+   - See `references/article-curation-workflow.md` for the full process
+
+## Output Format
+
+**Manual invocation (user asks directly):** Display full digest with all 4 topic sections + content opportunities.
+
+**From start-my-day cron (07:00):** Return condensed list:
+```
+**📰 Morning Briefing:**
+🔵 AI: [N items] — top headline
+🟢 Marketing: [N items] — top headline
+🟡 SEO: [N items] — top headline
+🟣 Research: [N items] — top headline
+Full digest: [[YYYY-MM-DD-Digest]]
+```
+
+## Provider Fallbacks
+
+`web_extract` and `web_search` may be unavailable (no provider configured). **Primary approach is `curl` via terminal.** Only use `web_extract`/`web_search` if they return data.
+
+### Fetch commands (preferred — curl + terminal)
+
+```bash
+# RSS feeds: curl + parse XML inline
+curl -sL --max-time 15 "https://bullrich.dev/tldr-rss/ai.rss" | head -500
+curl -sL --max-time 15 "https://rss.beehiiv.com/feeds/2R3C6Bt5wj.xml" | head -500
+curl -sL --max-time 15 "https://www.marketingaiinstitute.com/blog/rss.xml" | head -500
+curl -sL --max-time 15 "https://blog.hubspot.com/marketing/rss.xml" | head -500
+curl -sL --max-time 15 "https://searchengineland.com/feed" | head -500
+curl -sL --max-time 15 "https://www.searchenginejournal.com/feed/" | head -500
+curl -sL --max-time 15 "https://ahrefs.com/blog/feed/" | head -500
+curl -sL --max-time 15 "https://www.semrush.com/blog/feed/" | head -500
+curl -sL --max-time 15 "https://feedpress.me/mozblog" | head -500
+curl -sL --max-time 15 "https://neilpatel.com/blog/feed/" | head -500
+
+# arXiv API (XML — parse <entry> elements)
+curl -sL --max-time 15 "https://export.arxiv.org/api/query?search_query=all:marketing+AND+all:AI&sortBy=submittedDate&sortOrder=descending&max_results=10" | head -300
+
+# For truncated large feeds, extract titles only:
+curl -sL --max-time 15 "https://rss.beehiiv.com/feeds/2R3C6Bt5wj.xml" | grep -o '<title>[^<]*</title>' | head -15
+```
+
+## Source Health Notes
+
+| Source | Status | Note |
+|--------|--------|------|
+| TLDR AI | ✅ Active | `bullrich.dev/tldr-rss/ai.rss` — reliable, rich descriptions |
+| The Rundown AI | ✅ Active | `rss.beehiiv.com` — HTML entities in titles (`&#39;`), use grep for title extraction if truncated |
+| Marketing AI Institute | ✅ Active | Blog posts weekly, HubSpot-hosted |
+| HubSpot Marketing | ✅ Active | Very large feed, head -500 sufficient for recent items |
+| Search Engine Land | ✅ Active | WordPress RSS, standard format |
+| Search Engine Journal | ✅ Active | WordPress RSS, frequent updates |
+| Ahrefs Blog | ✅ Active | Fresh content, enterprise SEO focus |
+| Semrush Blog | ✅ Active | Data-driven SEO, frequent posts |
+| Moz Blog | ✅ Active | **Fixed in June 2026** — URL changed from `moz.com/blog/feed` → `feedpress.me/mozblog` |
+| Neil Patel | ✅ Active | Marketing + SEO insights, daily posts |
+| Google Search Central | ❌ Permanently broken | `developers.google.com/search/blog/feed` 404 — blog discontinued/unhosted as of 2026 |
+| arXiv API | ✅ Active | XML Atom feed, parse `<entry>` elements |
+| Web Search | ⚠️ Config-dependent | Requires `web.search_backend` configured via `hermes tools` |
+
+## Cron Mode Constraints
+
+- **`execute_code` is BLOCKED in cron mode.** Use individual `terminal` calls instead. Batch parallel `curl` fetches as separate tool calls in one response.
+- **`web_search` and `web_extract` may be unavailable** if not configured at the profile level.
+- When a tool fails with "No provider configured", silently fall back to `curl` + `terminal` — do not abort.
+
+## Content Creation Bridge
+
+After generating the digest, the user may ask to write articles from the news items. When doing so, follow the **Research-First** approach documented in `references/news-to-article.md`.
+
+### Research-First: News → Article Pipeline
+
+1. **Identify the source articles** linked in each news item — don't write from the digest summary alone
+2. **Read the original source articles** — use `curl -sL --max-time 20` with `python3 -c` for HTML text extraction; fall back to `browser_navigate` for Cloudflare/Vercel-protected sites
+3. **Extract key data**: specific numbers, quotes, mechanisms, and causal relationships
+4. **Synthesize multiple sources** into a coherent single narrative — don't just restate one article
+5. **Write in Thai** with:
+   - 📌 "Takeaway" callout boxes per section (labeled by topic, e.g. "GEO Takeaway", "Content Strategy Takeaway")
+   - Data tables with comparison columns and emoji indicators (✅/❌/🟢/🟡/🔴)
+   - A summary table of actionable takeaways at the end (with bulletproof reasoning column)
+   - Source citations at the bottom linking back to originals
+6. **Save article** to `~/vault/60_Articles/<slug>/article.md` with title/description/date/category/tags in frontmatter
+7. **Save brief** — optionally create a brief.md in the same 60_Articles/<slug>/ directory with the research notes and source summaries
+
+### Research fallback handling
+
+When `web_extract` and `web_search` have no provider configured (common in this profile):
+
+- **Generic HTML articles**: `curl -sL --max-time 20 "<URL>" | python3 -c "import sys,re; html=sys.stdin.read(); html=re.sub(r'<script[^>]*>.*?</script>', '', html, flags=re.DOTALL); html=re.sub(r'<style[^>]*>.*?</style>', '', html, flags=re.DOTALL); text=re.sub(r'<[^>]+>', ' ', html); text=re.sub(r'\s+', ' ', text).strip(); print(text[:15000])"`
+- **Sites with article tag**: Target `<article>` element specifically: `re.search(r'<article[^>]*>(.*?)</article>', html, re.DOTALL)` then strip tags
+- **Cloudflare/Vercel blocked sites**: Fall back to `browser_navigate` + `browser_snapshot(full=true)` or browser_scroll for longer articles
+- **VentureBeat**: Known to have Vercel security checkpoint; always use browser tools (curl won't work)
+- See `orbites-ai-products` skill for the HN Algolia API and GitHub Search API curl patterns
+
+### Thai content conventions
+
+- Title: Hook-driven, ≤ 60 chars, includes curiosity gap or contrarian claim
+- Opening paragraph: Hook that breaks reader expectation (not throat-clearing)
+- Structure per section: Data table → explanation paragraph → 📌 Takeaway box
+- End with: "ที่มา:" section listing all source articles as markdown links
+- Avoid: generic "ในโลกดิจิทัลวันนี้" openings, artificial FAQ, keyword stuffing
+
+## Error Handling
+
+- One source down: Skip, note in section header "(1 source unavailable)"
+- All sources in a topic down: Show "⚠️ No new items today" for that section
+- All sources down: Use yesterday's digest with warning
+- arXiv returns empty: Skip research section gracefully
+- Web search rate-limited: Use cached web results or skip
+- Terminal output truncated (>50KB): Use `grep` to extract key fields (titles, links)
+- Google Search Central 404: Skip silently, note in source summary
+- Moz stale: Skip entirely unless user asks for historical SEO content
--- a/skills/orbitos/orbites-ai-newsletters/references/article-curation-workflow.md
+++ b/skills/orbitos/orbites-ai-newsletters/references/article-curation-workflow.md
@@ -0,0 +1,64 @@
+# Article Curation Workflow — Daily Digest → 30_Research/
+
+## Why
+
+The daily digest generates a raw dump in `50_Resources/Newsletters/YYYY-MM/Raw/`.
+The user confirmed that high-value articles should **not** stay in Raw/ — they belong in `30_Research/` as permanent reference files.
+
+User's exact words: *"จำเรื่องนี้ไว้เลยนะ"* (remember this). The confirmed flow:
+
+```
+Raw/ (temp cache)
+  ↓ select high-value articles
+30_Research/<slug-title>.md  (permanent)
+  ↓
+Referenced from future digest or content planning
+```
+
+## Selection Criteria
+
+From the digest's Content Creation Opportunities section or notable articles:
+
+- High content creation potential (LinkedIn post, blog, tutorial)
+- Novel research or data-driven insights
+- Actionable for the user's SEO/marketing work (brand visibility, AI search, GEO)
+- Topic fits the user's niche (AI × Marketing, SEO, agentic systems)
+
+## File Format
+
+```markdown
+---
+type: article
+source: https://example.com/article
+source_digest: "[[50_Resources/Newsletters/YYYY-MM/YYYY-MM-DD-Digest]]"
+captured: YYYY-MM-DD
+tags: [article, ai-marketing, brand-visibility]
+---
+
+# [Article Title]
+
+## Key Takeaways
+- ...
+- ...
+
+## Why This Matters for Moreminimore
+- ...
+- ...
+
+## Action Items
+- [ ] Consider for LinkedIn post
+- [ ] Reference in [project name] content plan
+```
+
+## Cleanup
+
+- Remove the source raw file from `Raw/` after migrating
+- If multiple articles from same raw file, remove after last migration
+- Keep the digest itself (it's the curated summary, not the cache)
+
+## Example
+
+Article `chatgpt-opens-ads-all.md` was moved from:
+```
+50_Resources/Newsletters/2026-06/Raw/ → 30_Research/chatgpt-opens-ads-all.md
+```
--- a/skills/orbitos/orbites-ai-newsletters/references/news-to-article.md
+++ b/skills/orbitos/orbites-ai-newsletters/references/news-to-article.md
@@ -0,0 +1,138 @@
+# News-to-Article Workflow Reference
+
+Full workflow for taking news items / content ideas from a digest and writing in-depth articles.
+
+## Article Directory Structure
+
+```
+~/vault/60_Articles/
+├── YYYY-MM-DD-<slug>/
+│   ├── article.md       # Full article with frontmatter (title, description, date, category, tags, status: draft)
+│   ├── brief.md         # Optional: research notes, source summaries, angles considered
+│   └── images/          # Featured + inline images
+│       └── featured.png
+└── YYYY-MM-DD-<slug2>/
+    ...
+```
+
+## Article Frontmatter Template
+
+```yaml
+---
+title: "Hook-driven title (≤ 60 chars, Thai)"
+description: "SEO meta description (120-160 chars)"
+slug: "yyyy-mm-dd-kebab-case-slug"
+date: YYYY-MM-DD
+category: SEO | AI | Marketing | Research
+author: Macky
+tags: [Tag1, Tag2, Tag3, Tag4]
+status: draft
+---
+```
+
+## Research Phase (do this first, before writing)
+
+### Step 1: Identify source articles
+From the digest's news items, find the original source URLs. Never write from the digest summary alone — always read the originals.
+
+### Step 2: Read each source article
+Use this curl technique for most sites:
+
+```bash
+curl -sL --max-time 20 "<URL>" | python3 -c "
+import sys, re
+html = sys.stdin.read()
+# Strip scripts and styles
+html = re.sub(r'<script[^>]*>.*?</script>', '', html, flags=re.DOTALL)
+html = re.sub(r'<style[^>]*>.*?</style>', '', html, flags=re.DOTALL)
+# Try to get <article> content first
+match = re.search(r'<article[^>]*>(.*?)</article>', html, re.DOTALL)
+if match:
+    text = re.sub(r'<[^>]+>', ' ', match.group(1))
+else:
+    text = re.sub(r'<[^>]+>', ' ', html)
+text = re.sub(r'\s+', ' ', text).strip()
+print(text[:15000])
+"
+```
+
+For Cloudflare-protected sites (e.g. VentureBeat):
+- Fall back to `browser_navigate` → `browser_snapshot(full=true)` → `browser_scroll` as needed
+
+### Step 3: Extract key data per source
+For each source, extract:
+- **Numbers** — all specific metrics, percentages, dollar amounts
+- **Quotes** — direct statements from researchers/executives
+- **Mechanisms** — how something works (the causal chain)
+- **Contrasts** — before/after, old/new, claimed vs actual
+
+### Step 4: Synthesize across sources
+Organize findings into a coherent narrative:
+1. What's the core insight/problem each article describes?
+2. How do the sources overlap, complement, or contradict each other?
+3. What's the single actionable takeaway that readers can use today?
+
+## Writing Phase
+
+### Thai Content Structure (per article)
+
+```
+# [Hook-driven title]
+
+> [Hook paragraph — 1-2 sentences, breaks reader expectation]
+
+## 📚 สารบัญ
+- [Section 1]
+- [Section 2]
+- ...
+
+## [Section 1 — Core insight]
+[Data table with comparison columns]
+[Explanation paragraph — 2-4 sentences]
+📌 [Topic] Takeaway: [Actionable insight in 1-2 sentences]
+
+## [Section 2 — Mechanism]
+[Data table with before/after]
+[Explanation]
+📌 [Topic] Takeaway: [...]
+
+...
+
+## สรุป Actionable Takeaways
+| สิ่งที่ต้องทำ | เหตุผล |
+|:---|:---|
+| ✅ [Action] | [Why — bulletproof reasoning] |
+| ❌ [Avoid] | [Why — the trap to dodge] |
+
+---
+
+**ที่มา:**
+- [Author], "[Title]" — [Publication], [Date]
+```
+
+### Thai Writing Conventions
+
+| Element | Convention |
+|:---|:---|
+| **Title** | Hook-driven ≤ 60 chars. Curiosity gap or contrarian claim. Include target keyword naturally |
+| **Opening** | Break expectation — "ถ้าคุณคิดว่า X — คุณคิดผิด" / "ไม่ใช่แค่ X — มันคือ Y" |
+| **Section labels** | 📌 "GEO Takeaway" / "Content Strategy Takeaway" / "Business Takeaway" |
+| **Tables** | Comparison columns with emoji indicators (✅ ❌ 🟢 🟡 🔴 🏆) |
+| **Code/terms** | Use backticks for English technical terms (`result_source`, `turn_use_case`) |
+| **Numbers** | Bold the headline number — "เร่งความเร็วได้ **85%**" |
+| **Sources footer** | "**ที่มา:**" section listing all source articles as markdown links |
+
+### Avoid
+- Generic openings: "ในโลกดิจิทัลวันนี้", "ด้วยความก้าวหน้าของเทคโนโลยี"
+- Artificial FAQ sections — don't add them unless the content genuinely raises questions
+- Keyword stuffing — use Thai synonyms naturally
+- Over-explaining — trust the reader to connect simple dots
+
+## Pitfalls
+
+1. **web_extract / web_search may not be configured** — always have curl+browser fallback ready. Script/security warnings from pipe-to-python are a real risk; get auto-approval or use inline parsing.
+2. **Majority of articles are in the 60_Articles/<slug>/article.md format** — check existing structure before saving. Never invent a new format.
+3. **Source article may be behind paywall or bot-block** — curl can't read everything. Browser tools handle Cloudflare/Vercel. YouTube and social platforms need different approaches.
+4. **Content date ≠ publish date** — use the source research date or the article's original publication date, not today's date, for the article slug and frontmatter.
+5. **Multiple articles from one digest session** — batch them in parallel via delegate_task if independent; write sequentially if they share data.
+6. **arXiv API returns Atom XML, not RSS** — parse `<entry>` elements specifically; the `head -300` approach in the newsletter skill truncates at a fixed line count, so use `python3 -c` for proper XML parsing.