Forum Monitor Skill | Agent Skills

Forum Monitor

Overview

Systematic workflow for monitoring online communities to discover recurring user pain points. Transforms unstructured forum discussions into ranked, evidence-backed opportunity reports suitable for automated app generation pipelines.

Core principle: Pain points with high frequency and high engagement are the strongest product signals. Monitor, classify, rank, report.

When to Use

Discovering product opportunities from community feedback
Running periodic (daily/weekly) market research scans
Identifying trending complaints or feature requests
Competitive intelligence gathering from user discussions
Feeding the app-generation-workflow with validated pain points

Cron Integration

This skill is designed to run on a schedule via CronCreate:

CronCreate({
  name: 'weekly-forum-scan',
  schedule: '0 9 * * MON', // Every Monday at 9 AM
  prompt:
    'Run forum monitor scan for [target domain]. Invoke Skill({ skill: "forum-monitor" }). Save report to .claude/context/reports/backend/',
});

Workflow

Step 1: Configure Target Forums

Define the forums to monitor based on the target domain:

Forum Selection Matrix:

| Forum | Best For | Query Pattern | | ------------- | ------------------------------- | --------------------------------------------- | | Reddit | Consumer pain points, UX issues | site:reddit.com <topic> frustrating OR wish | | Hacker News | Developer tools, B2B SaaS | site:news.ycombinator.com <topic> | | ProductHunt | New product gaps, feature envy | site:producthunt.com <topic> | | Indie Hackers | Solo dev pain points, pricing | site:indiehackers.com <topic> | | Dev.to | Developer workflow friction | site:dev.to <topic> pain OR annoy |

Query Templates:

"{topic} frustrating OR annoying OR wish OR need OR missing"
"{topic} alternative to OR better than OR looking for"
"{topic} feature request OR roadmap OR please add"

Step 2: Scrape and Collect

For each configured forum, execute searches and extract content:

// Search for pain points
WebSearch({ query: 'site:reddit.com {topic} frustrating OR wish OR need 2026' });

// Fetch specific threads with high engagement
WebFetch({
  url: '{thread-url}',
  prompt:
    'Extract all complaints, feature requests, and pain points. For each, note the exact quote, upvote count, and whether others agreed.',
});

Collection Requirements:

Minimum 20 posts per forum per scan
Include posts from the last 30 days (or configurable window)
Capture: title, URL, community, engagement (upvotes + comments), date, key quotes

Step 3: Classify Pain Points

Categorize each collected item into one of these categories:

| Category | Signal Words | Example | | ------------------- | --------------------------------------- | ------------------------------------ | | missing-feature | "wish", "need", "please add", "roadmap" | "I wish Notion had offline mode" | | workflow-friction | "slow", "clunky", "takes forever" | "It takes 10 clicks to export a PDF" | | bug-report | "broken", "crashes", "error" | "The app crashes on large files" | | pricing | "expensive", "not worth", "free alt" | "Too expensive for a solo dev" | | ux-confusion | "confusing", "can't find", "intuitive" | "I had no idea where settings were" | | integration-gap | "connect to", "integrate with", "API" | "No Zapier integration available" |

Step 4: Cluster and Deduplicate

Group similar pain points into clusters:

Exact duplicates: same complaint, different posts -> merge, sum engagement
Semantic duplicates: similar complaint, different wording -> cluster, note variants
Related but distinct: same domain, different problems -> keep separate

Step 5: Rank by Opportunity Score

For each cluster, compute:

Opportunity Score = (Frequency x 0.4) + (Engagement x 0.3) + (Recency x 0.2) + (Sentiment Intensity x 0.1)

Where:

Frequency: Number of unique posts mentioning this pain point (normalized 0-10)
Engagement: Total upvotes + comments across all posts (normalized 0-10)
Recency: How recent the complaints are (last 7 days = 10, last 30 days = 5, older = 2)
Sentiment Intensity: How strongly negative the language is (0-10)

Step 6: Generate Report

Write structured output to .claude/context/reports/backend/forum-monitor-report-{YYYY-MM-DD}.md:

<!-- Agent: forum-monitor-agent | Task: #{id} | Session: {date} -->

# Forum Monitor Report

**Scan Period**: {start-date} to {end-date}
**Forums Monitored**: {list}
**Total Posts Analyzed**: {count}
**Pain Point Clusters Found**: {count}

## Top Pain Points (Ranked by Opportunity Score)

| Rank | Pain Point | Category | Freq | Engagement | Score | Sources |
| ---- | ---------- | -------- | ---- | ---------- | ----- | ------- |
| 1    | [desc]     | [cat]    | [n]  | [n]        | [n.n] | [n]     |

## Detailed Findings

### 1. [Pain Point Title] (Score: X.X)

**Category**: [type]
**Frequency**: [n] mentions across [n] sources
**Engagement**: [total upvotes] upvotes, [total comments] comments
**Forums**: [list of forums where this appeared]

**Representative Quotes:**

1. "[exact quote]" - [source URL] ([n] upvotes)
2. "[exact quote]" - [source URL] ([n] upvotes)
3. "[exact quote]" - [source URL] ([n] upvotes)

**App Opportunity Assessment:**

- Buildable as standalone app: YES/NO
- Estimated complexity: LOW/MEDIUM/HIGH
- Existing solutions: [list or "none found"]
- Differentiation angle: [what would make a new solution win]

Iron Laws

ALWAYS cite source URLs for every finding -- unverifiable claims are worthless
NEVER fabricate engagement metrics -- counts must come from actual collected data
ALWAYS cross-reference at least 2 forums before declaring a trend
NEVER include content from private or gated forums -- public content only
ALWAYS include verbatim quotes -- user language is more valuable than agent summaries

Anti-Patterns

| Anti-Pattern | Why It Fails | Correct Approach | | ----------------------------- | --------------------------------------------------- | --------------------------------------------- | | Single-forum reports | One community is not representative | Cross-reference 2+ forums before trending | | Subjective ranking | Personal opinion is not data | Use weighted opportunity formula | | Missing source URLs | Downstream agents cannot validate findings | Every finding must have a clickable source | | Stale data without date range | Trends from 2024 are not 2026 trends | Always specify scan period in report header | | Over-counting duplicates | Same user posting in 3 threads is not 3 data points | Deduplicate by unique user + unique complaint |

Related Skills

browser-automation -- for deeper scraping when WebSearch/WebFetch are insufficient
deep-research -- for comprehensive investigation of specific pain points
feedback-analysis -- for structured sentiment and NPS analysis

Assigned Agents

| Agent | Role | | --------------------- | --------------------------------------- | | forum-monitor-agent | Primary -- executes the full workflow | | researcher | Supporting -- deeper investigation | | app-generator-agent | Consumer -- reads reports for app ideas |

Memory Protocol (MANDATORY)

Before starting:

node .claude/lib/memory/memory-search.cjs "forum monitor pain points trends"

Read .claude/context/memory/learnings.md

After completing:

New monitoring pattern -> .claude/context/memory/learnings.md
Forum access issue -> .claude/context/memory/issues.md
Scoring model decision -> .claude/context/memory/decisions.md

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.

Agent Skills: Forum Monitor

Install this agent skill to your local

Skill Files