trending-now（现在流行）

Setup

On first use, read setup.md and lock integration behavior before starting trend monitoring.

When to Use

User needs ongoing updates about what is trending across the internet and social platforms, with special attention to X and rapid shifts in conversation. Use this skill to define topic watchlists, run heartbeat-based research cycles, rank signal strength, and send concise messages only when there is meaningful change.

Architecture

Memory lives in ~/trending-now/. See memory-template.md for the baseline structure.

~/trending-now/
|-- memory.md                 # Activation behavior, scope, and monitoring preferences
|-- topics.md                 # Active topics, query variants, and relevance boundaries
|-- runs.md                   # Heartbeat run history and change detection summary
`-- alerts.md                 # Alerts sent, impact notes, and false-positive log

Quick Reference

Use the smallest relevant file for the current task.

Requirements

• Web access for live trend validation.
• User-approved scope for topics, geographies, and languages.
• Timezone and active hours for heartbeat delivery behavior.

Never claim a trend is current without timestamped evidence from at least two independent sources.

Data Storage

Local notes in ~/trending-now/ include:

• monitored topics with query variants and stop words
• run-level evidence links and freshness timestamps
• alert history with confidence and post-send outcomes
• rejected spikes and false-positive rationale

Core Rules

1. Define Topic Scope Before Monitoring

Each topic must include:

• explicit intent (brand, product, industry, culture, or breaking-event)
• inclusion and exclusion criteria
• audience and geography boundaries

Without scope, trend monitoring becomes noisy and low trust.

2. Use HEARTBEAT.md as the Operating Contract

Always maintain topic and output rules in HEARTBEAT.md. Every cycle must follow one contract:

• actionable update -> send structured message
• no meaningful change -> return HEARTBEAT_OK

Do not send filler summaries when there is no decision-relevant movement.

3. Prioritize Source Diversity with X as Fast Signal

For each topic, gather evidence from:

• X for velocity and narrative emergence
• at least one community source (Reddit, forums, niche communities)
• at least one publisher or search trend source

Single-network spikes are hypotheses, not confirmed trends.

4. Enforce Freshness and Recency Windows

Classify findings by age:

• hot: <= 6 hours
• recent: <= 24 hours
• stale: > 24 hours

Escalate only hot or recent signals unless the user explicitly requests longer-horizon analysis.

5. Score Trend Strength Before Alerting

Apply a fixed score per candidate trend:

• volume shift
• cross-source confirmation
• novelty versus prior runs
• user relevance
• action urgency

If score is below threshold, store in watchlist and do not alert yet.

6. Send Messages in Decision-Ready Format

Every alert message must include:

• what changed
• why it matters now
• confidence and risks
• one concrete next action

No long narrative dumps. Message length should fit quick mobile reading.

7. Protect Cost and Credibility

Start with low-cost checks, then deepen only when a signal passes threshold. Never use paid APIs every cycle unless the user explicitly approves budget. Always mark uncertain claims and avoid overconfident language.

Common Traps

• Treating repost volume on X as proof of broad trend adoption -> repeated false positives.
• Using only one source family -> hype detection without validation.
• Ignoring recency windows -> old stories presented as breaking updates.
• Sending alerts without an action recommendation -> interesting but not useful output.
• Expanding topic scope mid-cycle without user approval -> relevance drift.
• Logging conclusions without links and timestamps -> impossible to audit later.

External Endpoints

No other data should be sent externally unless the user explicitly approves additional sources.

Security & Privacy

Data that leaves your machine:

• topic keywords used for live trend research
• source lookups needed to verify recency and momentum

Data that stays local:

• monitoring preferences and topic definitions under ~/trending-now/
• run history, confidence scores, and alert outcomes

This skill does NOT:

• access private social accounts by default
• post content on any social platform automatically
• run undeclared external requests outside approved sources

Trust

This skill relies on public internet and social sources, including X and other platforms the user approves. Only install and run it if you trust those sources and the external services used for research.

Related Skills

Install with clawhub install <slug> if user confirms:

• monitoring - Build monitoring loops with clear thresholds and escalation paths
• news - Structure news tracking and summarization for time-sensitive decisions
• competitor-monitoring - Track competitor moves with disciplined evidence and update cadence
• in-depth-research - Expand weak or mixed signals into deeper evidence-backed analysis
• digest - Turn many raw updates into compact, high-signal briefings

Feedback

• If useful: clawhub star trending-now
• Stay updated: clawhub sync