SkillSmith Social Narrative Tracker
A sentiment score is a symptom; the narrative is the diagnosis. This Skill ships the attributed narratives behind the number — not the number alone.
What it does
Most "social listening" output is a sentiment percentage and a word cloud — which hides the only things a decision-maker can act on: which stories a community is telling, where they live, and what is new since last time. This Skill turns one input — "track the social narrative around <topic / brand / $TICKER>" — into a structured, multi-source narrative brief.
It clusters real chatter into a small set of named narratives (bull/bear for a ticker, positive/negative themes for a brand), gives a per-platform sentiment lean, surfaces emerging concerns, contrasts how Reddit, X, YouTube and Hacker News differ, and pulls representative quotes that each carry a source locator. Every run is research-grade: nothing is fabricated, and thin coverage is reported honestly rather than backfilled.
How it works
A 6-phase workflow, each with a hard gate before the next:
| Phase | Output |
|---|---|
| 1. Scope & disambiguate | Lock the exact subject, angle, platform set; resolve name collisions; pick evidence mode (live search / pasted / both) |
| 2. Gather (attributed) | Targeted per-platform queries; capture platform + locator + date + verbatim quote for every citable item |
| 3. Cluster into narratives | Group evidence into 3–6 named narratives with prevalence, sentiment lean, and representative quotes |
| 4. Cross-platform & emergence | Contrast how platforms differ; flag small-but-accelerating signals as EARLY/UNCONFIRMED |
| 5. Diff & metrics | "What changed" vs a prior brief; define trackable indicators for next period |
| 6. Assemble & guardrail-check | Emit the fixed brief; verify attribution, anti-fabrication, and the right disclaimer |
⭐ Featured example — tracking the Arc Browser narrative
Input: "Track the social narrative around Arc Browser (The Browser Company) — reception of the maintenance-mode + Dia/Atlassian pivot."
Output: a multi-source brief that names the dominant narratives, leans sentiment per platform, and backs every quote with a real locator — built from a live run over Hacker News and community reviews:
![Social narrative brief: executive read, per-platform sentiment lean, and named narratives with attributed quotes]
It then separates the platforms instead of averaging them — Reddit's emotional betrayal vs Hacker News' product skepticism — and lists representative quotes each tied to a source, plus emerging signals and what to track next:
![Cross-platform differences and source-attributed representative quotes, with emerging signals and trackable indicators]
The quotes are real and traceable — e.g. "I want a browser in my apps, not apps in my browser" from Hacker News thread 45126358 (523 points, 505 comments), and review-quoted community lines — never invented. Where live coverage was thin (X), the brief says so and asks for pasted posts rather than fabricating chatter.
Why it's different
- Narratives, not a number. The market already has tools that spit out a sentiment %. The product here is the structured, named narrative set with attributed evidence — the diagnosis behind the score.
- Source attribution is enforced. Every representative quote carries a platform + locator + date. No locator → the quote is cut. That single rule is what separates research from a vibe.
- Anti-fabrication by design. Thin or blocked coverage is reported honestly and the buyer is asked to paste source material — the Skill never characterizes "the general vibe" from memory or invents quotes/stats.
- Cross-platform separation. Reddit, X, YouTube and HN are read in their own dialects and contrasted, not blended into one misleading average.
Live-coverage note (honest by design): Hacker News and open forums with stable permalinks are the most reliably reachable via live search; Reddit and X first-hand posts are often not retrievable by a general web search, so for those the Skill asks you to paste the threads rather than guess. Treat it as "live where reachable, paste-assisted for Reddit/X" — not a fully-automatic four-platform crawler. Media retellings of a post are marked
[secondary], never passed off as a first-hand quote.
- Built to track over time. A "what changed" diff and trackable indicators make successive briefs comparable — the recurring value behind a subscription.
Disclaimer / Scope
This Skill produces a synthesis of publicly expressed community opinion for informational and research purposes only. It reflects what people are saying, not verified fact, and is not professional, legal, PR, or investment advice. Sentiment leans describe the cited sample, not a complete census. For any stock/ticker subject it adds an explicit "not investment advice" disclaimer. It uses public content only — no private data, no de-anonymization, no surveillance of individuals — and labels unverified claims as UNCONFIRMED rather than amplifying them.
