trainingsites/campus-ai-os/content-scanner

content-scanner — find the evidence

Discovers and signal-extracts content/strategy opportunities. Hands ranked findings to scoring-engine, then brief-builder.

Kernel preamble §4 applies: references/kernel-preamble.md.

CRITICAL DIRECTIVES

• Tool-agnostic. Baseline tier (WebSearch + WebFetch) always works; premium and internal sources light up ONLY if detected live. Never assume a connector; never prompt to install one.
• Zero hardcoded paths. If a fetch returns an empty shell, ask for a paste — never fabricate findings.
• Respect a research mandate if one exists (strategy-council may have written mandate.json with hunt_for/avoid).

Step 1 — Mode & inputs

• /atlas-scan (quick): top 3 opportunities, ~2 min. /atlas-daily (full): top opportunities + full signal detail. /atlas-validate [topic]: test one specific topic.
• Read niche/business_model/competitors from the resolved context (CONTEXT_CANON (.campus-os/context-files.md, else the plugin's templates/context-files.md): icp, offers) or team-private atlas/config/niche.md — never hard-code shared/*. Load any mandate.json.

Step 2 — Pull sources (see references/sources.md)

Always on: Google SERP, Reddit, YouTube, Quora/PAA, top-3 competitor sites, YouTube comment mining, 1–3★ review mining. Premium (if present): deep Reddit/YouTube via Apify, multi-page crawls via Firecrawl. NEW capability-gated (only if the tool/data is live): sales & support mining (top signal), voice-of-customer polls, trend/seasonality, social listening, SEO volume, answer-engine (AEO) monitoring. Prioritize the user's configured `sources` (where to research) from `atlas/config/niche.md` — include and weight those first, then fill with the always-on set. Weight by business_model.

Expert-tools check (one line, before pulling). Ask what a top researcher in this niche would use for this question that isn't connected here (a keyword-volume tool, a review database, the owner's own sales or support data, a poll). Say it in one line ("Expert-tools check: a pro would also use … — not connected, so demand is read from search and comments only") and list it in the report as a known gap. Never install or sign up for anything; this only names the gap so the score is honest about what it couldn't see.

Step 3 — Extract signals

For each candidate topic capture evidence for the four signals: demand (who's asking, volume), gap (saturation/what's missing), timing (rising/seasonal), fit (matches audience + offers). Note which internal buckets apply (references/capability-tiers.md) for the 1.5× multiplier. Check the already-covered guard (published-content.md / Content bucket).

Step 4 — Hand off

Pass candidates + evidence to scoring-engine. For /atlas-daily, continue to brief-builder for the top picks. For /atlas-validate, return one topic's signal read + a go / refine / skip call.

Step 5 — Report

Write a scan report to the handoff location's reports dir. List ranked opportunities with one-line evidence each, and the expert-tools gap from Step 2. Offer: "Want briefs built for the top picks? (/atlas-daily)"