Newsletter

Newsletter Curation Skill

Fetches and curates Gmail newsletter content by label using the gws CLI Python script, classifies messages by topic, and produces structured data for the /newsletter command to generate a premium intelligence digest report.

Prerequisites

• gws CLI authenticated (gws auth login)
• uv available in PATH
• gmail_fetch_newsletter.py script available via plugin-cache path resolution

---

Procedure

Step 1 — Resolve Script Path

_S=$(find "$HOME/.claude/plugins/cache" -name "gmail_fetch_newsletter.py" -path "*/claude-skills/*" 2>/dev/null | sort -rV | head -1)
[ -z "$_S" ] && _S="scripts/gmail_fetch_newsletter.py"

Step 2 — Resolve Label

If the user passes a label name (not an ID starting with Label_), resolve it:

uv run python "$_S" --label-id dummy --list-labels 2>/dev/null \
  | jq '[.[] | select(.type=="user") | {id, name}]'

Match the user-supplied string against name (case-insensitive, partial match OK). If no match, abort with a message listing available user labels.

If the argument already looks like a label ID (Label_\d+ or Label_[a-f0-9]+), use it directly.

Step 3 — Fetch Messages

uv run python "$_S" \
  --label-id "$LABEL_ID" \
  --days "$DAYS" \
  --max-results 40 \
  2>/dev/null

Output: JSON array of {id, from, subject, date, snippet, body_text, links}.

If empty array → report "No messages found for this label in the specified time range." and stop.

Step 4 — Classify Messages by Topic

For each message, analyze from + subject + snippet + body_text and assign one of:

| Category | Key Signals | |----------|-------------| | AI & Engineering | LLM, GPT, Claude, Gemini, Copilot, AI, ML, model, coding, developer, deep learning, RAG, agent, vibe coding, ChatGPT, Anthropic, OpenAI, inference, reasoning | | Big Tech & Investment | Google, Apple, Microsoft, Meta, Amazon, NVIDIA, big tech, stock price, investment, M&A, market cap, earnings, funding, valuation | | Startup & Product | startup, launch, funding, Series A/B/C, ProductHunt, new service, beta, Kickstarter | | Tools & Infrastructure | Railway, Docker, K8s, AWS, GCP, Azure, Slack, GitHub, DevOps, infrastructure, CI/CD, SDK | | Other | everything else |

Each message gets exactly one category (best match wins; AI > BigTech > Startup > Tools > Other).

Step 5 — Extract Keywords for Word Cloud

From all subject + snippet text, extract:

• Top 20 significant nouns/terms (exclude stopwords: the, a, an, is, in, of, to, for, and, or, with, your, our, this, that)
• Group by category
• Output as nested list for Mermaid mindmap

Format for mindmap:

      AI_KEYWORD_1
      AI_KEYWORD_2

(8 spaces + term, under the parent category branch)

Step 6 — Build Gantt Milestones

For messages with specific event dates (e.g., "May 27", deadline mentions):

• Extract the date
• Create gantt task: Task_Name : milestone, YYYY-MM-DD, 0d
• Group under appropriate category section

For non-event messages, use dateOnly as a 1-day task span.

Step 7 — Format Each Item for T7 Template

For each classified message, produce a markdown block:

### N.M [Subject — cleaned up, no emoji spam]
**Source** · [Sender Name] · [Date dd MMM YYYY]

[2~3 sentence analytical summary — synthesize body_text + snippet.
Focus on what this means for the reader, not just what it says.
Use analyst voice: "what this signals is", "the key takeaway is", "from an engineer's perspective"...]

**Why it matters**: [1 sentence takeaway in bold]

**Links**:
- [Link 1 Label](URL)
- [Link 2 Label](URL)   ← include only meaningful links (2~3 max per item)

---

Quality rules:

• Never just copy the snippet verbatim — synthesize
• For AI items: connect to practical developer use
• For BigTech: connect to market/investment implications
• For Startups: note the "so what" for engineers or investors
• Skip links that look like tracking pixels or unsubscribe URLs

Step 8 — Compute Aggregates

• TOTAL_MESSAGES: total count
• Category counts: AICOUNT, BIGTECHCOUNT, STARTUPCOUNT, TOOLSCOUNT, OTHER_COUNT
• TOP_SENDERS: top 3 sender domains (e.g., "Railway, Substack, IITP")
• WEEK_LABEL: e.g., "Week of May 19"
• START_DATE, END_DATE: actual date range of fetched messages
• KEY_INSIGHT_1~3: one-sentence summary of the most important finding per top category

Step 9 — Load T7 Template

Resolve template path:

_TPL=$(find "$HOME/.claude/plugins/cache" -path "*/claude-skills/*/templates/research" -type d 2>/dev/null | sort -rV | head -1)
[ -z "$_TPL" ] && _TPL="templates/research"

Read $_TPL/T7-newsletter-curation.md and replace all [[PLACEHOLDER]] values with computed data from Steps 4–8.

Output the filled report string to the /newsletter command for file saving.

---

Section Placement Priority

1. 🤖 AI & Engineering — highest priority for AI-focused engineers. Each item includes "engineer's perspective" insight.
2. 📈 Big Tech & Investment — big tech strategic shifts and market-moving events. Add investor-angle commentary.
3. 🚀 Startup & Product — notable new products and funding. Add "why you should try this" one-liner.
4. 🔧 Tools & Infrastructure — dev tools and infra updates.
5. 📋 Other

Empty Section Handling

If a section has 0 items: > No relevant content in this period. — one line only.

Link Filtering Rules

• Exclude tracking/unsubscribe URLs (track., pixel., unsubscribe, etc.)
• Max 3 links per item
• Use meaningful labels for link text, not raw URLs

---

Mermaid Guidelines

English labels only in diagrams — no non-ASCII characters in node labels. For the mindmap, use short English/abbreviated terms only.

Example mindmap leaf expansion:

    🤖 AI & Dev
      LLM
      Claude
      Vibe-Coding
      n8n

For the gantt, use dateFormat YYYY-MM-DD and axisFormat %m/%d. For pie chart, use integer values only.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: liks79
• Source: liks79/claude-skills
• License: MIT
• Homepage: https://github.com/liks79/claude-skills

Install and usage instructions live in the source repository linked above.