weekly-sentiment-summary

# Weekly Sentiment Summary Bigger-picture view for growth, product, and devex teams. Typically run Monday morning covering the prior week. Uses the **Octolens** MCP server. ## Workflow ### Step 1: Determine date ranges - **This week**: the most recent completed Mon–Sun (or the user-specified range) - **Prior week**: the 7 days immediately before this week ### Step 2: Fetch mentions for both weeks 1. Call `list_mentions_context` to get available keyword IDs and filter syntax. 2. Identify your product's keywords (look for keywords matching your product name, domain, and any known brand handles). > **Setup required:** Configure your product's keyword identifiers in Octolens before using this skill. 3. For each week, query `list_mentions` with those keyword IDs, `relevance=[0]`, and the week's date range. Set `endDate` to the day after Sunday to ensure full coverage. 4. Set `includeAll: false` and `limit: 100`. 5. Paginate with `cursor` until exhausted or 500 mentions per week. ### Step 3: Aggregate both weeks For each week, count: - `pos`, `neu`, `neg` by sentiment - `total = pos + neu + neg` - Tag distribution (bug_report, user_feedback, competitor_mention, buy_intent, product_question, etc.) ### Step 4: Compute sentiment scores For each week: ``` sentiment_score = ((pos - neg) / total) * 100 ``` Range: -100 to +100. Round to nearest integer. Compute deltas: - `Δ volume = this_week_total - prior_week_total` - `Δ score = this_week_score - prior_week_score` Use `+` prefix for positive deltas, `-` for negative. Show as absolute numbers, not percentages. ### Step 5: Identify patterns by sentiment (this week only) Before analyzing, filter out: - **Employee posts** (replies from your company's team members to users) - **Spam / reseller posts** (e.g. discounted subscription sellers, engagement bait) - **Cross-post duplicates** (same content on Twitter and Bluesky — count once, prefer higher-reach version) For each sentiment category (Positive, Neutral, Negative), identify **3-5 key themes or patterns** from this week. Focus on: - What are users saying? What themes emerge? - Are there patterns across multiple mentions? - What's strategically meaningful vs. noise? Writing style: same as daily skill — lead with pattern/theme, include mention count, provide 1-2 representative links. **For Positive section only**: After pattern summaries, include 1-3 direct quotes that work as testimonials. Pick the most specific, enthusiastic quotes that show clear value. Format as: ``` _"[exact quote]"_ — <url|@username or source> ``` ### Step 6: Identify notable patterns Compare this week vs prior week. Look for: - **Recurring themes**: topics/features/complaints that appeared multiple times this week - **Trending tags**: tags with significant volume changes vs prior week - **New signals**: topics that appeared this week but not last week Max 5 patterns. Each should include a delta or comparison when possible (e.g. "AI agent mentions up from 3 → 12 this week"). ### Step 7: Product feedback patterns Scan this week's mentions tagged `user_feedback`, `bug_report`, `product_question` for **recurring product feedback themes**. Focus on specific features, workflows, or pain points mentioned multiple times. Look for: - Feature requests that appeared multiple times - Specific bugs or technical issues with volume - Pricing/credit feedback patterns - UI/UX feedback themes Summarize in 2–4 bullets with mention counts. Example: "Feature requests for BYOK on free plan (5 mentions)" or "UI lag complaints when long threads open (3 mentions)". ### Step 8: Write markdown report and create PR 1. Format the report as a markdown file (~500 words max) using the template below. 2. Save it to `reports/weekly_sentiment_analysis/weekly_sentiment_report_<end_date>.md` where `<end_date>` is the Sunday of the covered week in `YYYY-MM-DD` format. 3. Create a new branch, commit the report, push, and open a PR. Markdown template: ```markdown # Weekly Sentiment Summary — <start_date> → <end_date> **<total>** mentions (**<Δ>** vs last week) | 🟢 <pos> · ⚪ <neu> · 🔴 <neg> | Score: **<score>** (**<Δ>**) ## Positive - <pattern summary with count> — [link](url), [link](url) - <pattern summary with count> — [link](url) ... ### Testimonials > "<exact quote>" — [@username](url) > "<exact quote>" — [@username](url) ## Neutral - <pattern summary with count> — [link](url), [link](url) - <pattern summary with count> — [link](url) ... ## Negative - <pattern summary with count> — [link](url), [link](url) - <pattern summary with count> — [link](url) ... ## Notable Patterns - <pattern with week-over-week Δ> - <pattern with week-over-week Δ> ... ## Product Feedback Patterns - <specific product feedback theme with count> - <specific product feedback theme with count> ... ``` Omit any section that has no meaningful content. ## Edge Cases - **Low volume week** (<50 mentions): Still produce the report but note the low volume. Reduce highlight/lowlight counts proportionally. - **First run (no prior week data)**: Skip deltas and comparisons — just report this week's absolute numbers and note that week-over-week comparison will be available next week. - **Non-English mentions**: Include if noteworthy, with language/region context in the takeaway.