prod-trends

Trends Analysis — Weekly Comparison

Routine that compares community, GitHub, and financial metrics week-over-week to detect patterns, risks, and opportunities.

Always respond in English.

Data Sources

1. Community (Discord)

Read previous reports in:

• workspace/community/reports/daily/ — daily pulses (HTML)
• workspace/community/reports/weekly/ — weekly reports (HTML)

Extract from HTML or generate from data:

• Messages per day (volume)
• Active members (WAM)
• Unanswered questions
• Overall sentiment
• Top recurring topics

2. GitHub

Read previous reports in:

• workspace/projects/github-reviews/ — reviews (HTML)

Extract or generate:

• Open PRs (trend: accumulating or being resolved?)
• Open vs closed issues
• Stars/forks (growth)
• Commits per week (team activity)
• Average open PR time

3. Financial

Query data via skills:

• /int-stripe — MRR, cobranças, reembolsos, assinaturas ativas
• /int-omie — accounts receivable/payable (if available)

Metrics:

• MRR (Monthly Recurring Revenue)
• Monthly charges vs previous month
• Refunds
• Active subscriptions (growth/churn)

4. Operational (ADWs)

Read runner metrics:

• ADWs/logs/metrics.json — runs, success rate, avg time per routine

Workflow

Step 1 — Collect current week's data

Fetch the most recent data from each source (last 7 days).

Step 2 — Collect previous week's data

Fetch data from 7-14 days ago for comparison. If it does not exist (first run), mark as "baseline" and skip comparison.

Step 3 — Calculate trends

For each metric, calculate:

• Current vs previous value
• Absolute and percentage variance
• Direction: ↑ (rising), ↓ (falling), = (stable)
• Classification: 🟢 healthy, 🟡 attention, 🔴 risk

Classification criteria:

Step 4 — Detect patterns

Analyze recent weeks (as many as available) and identify:

• Persistent trends — metric rising/falling for 2+ consecutive weeks
• Correlations — e.g., increase in GitHub issues + increase in Discord questions = possible bug
• Anomalies — unusual spike or drop vs average
• Seasonality — recurring patterns (e.g., Monday has more activity)

Step 5 — Generate HTML report

Read the template at .claude/templates/html/custom/trends-report.html. Replace the placeholders {{...}} with the actual data.

Overall health classification:

• All 🟢 or mostly 🟢: healthy — "Healthy"
• Mix of 🟢 and 🟡: mixed — "Attention"
• Any 🔴: risk — "Risk"

REQUIRED: Always generate the HTML first. Read the template, replace the placeholders, and save the complete HTML file. This applies even on the first run (baseline) — even without comparison, fill the scorecard with current values and "—" for previous.

Save HTML to workspace/daily-logs/[C] YYYY-WXX-trends.html.

Then, also save a summarized markdown version to workspace/daily-logs/[C] YYYY-WXX-trends.md:

# Trends Analysis — Week {WXX}

## Executive Summary
{3 bullets: what improved, what worsened, opportunity}

## Scorecard

| Area | Metric | Current | Previous | Var | Trend | Status |
|------|---------|-------|----------|-----|-------|--------|
| Community | WAM | {N} | {N} | {+/-X%} | ↑/↓/= | 🟢/🟡/🔴 |
| Community | Unanswered questions | {N} | {N} | | | |
| Community | Sentiment | {label} | {label} | | | |
| GitHub | Open PRs | {N} | {N} | | | |
| GitHub | Unanswered issues | {N} | {N} | | | |
| GitHub | Stars (week) | {N} | {N} | | | |
| Financial | MRR | R${N} | R${N} | {var%} | | |
| Financial | Active subscriptions | {N} | {N} | | | |
| Operational | Success rate ADWs | {X}% | {X}% | | | |

## Detected Patterns
- {pattern 1 with evidence}
- {pattern 2 with evidence}

## Risks
- {risk with supporting metric}

## Opportunities
- {opportunity based on data}

## Recommendations
1. {concrete action based on data}
2. {concrete action}

Step 6 — Save snapshot

Save a snapshot of current metrics to memory/trends/YYYY-WXX.json to accumulate history:

{
  "week": "YYYY-WXX",
  "date": "YYYY-MM-DD",
  "community": {"wam": N, "messages": N, "unanswered": N, "sentiment": "positive"},
  "github": {"prs_open": N, "issues_open": N, "issues_unanswered": N, "stars_week": N, "commits_week": N},
  "financial": {"mrr": N, "subscriptions": N, "refunds": N},
  "operational": {"adw_runs": N, "adw_success_rate": N, "adw_avg_seconds": N}
}

Create memory/trends/ if it does not exist.

Rules

• First run = baseline — no comparison, just collect and save snapshot
• Real data — do not fabricate metrics, use what is available
• If a source has no data, skip — do not block due to a missing report
• Focus on action — each insight should lead to a concrete recommendation
• Do not alarm without evidence — red only when the metric truly indicates risk