Review Newsletter Subscriptions

review-newsletter-subscriptions

Audit and triage newsletter subscriptions by cross-referencing ingestion reports with user suggestion review history over the past 30 days (or a custom date range).

Workflow Overview

graph TD
    Trigger["User Request / Scheduled Audit"] --> RunScript["Step 1: Execute Telemetry Script<br/>(scripts/analyze_subscriptions.py)"]
    RunScript --> Triage["Step 2: Apply Multi-Dimensional Triage<br/>(references/triage_criteria.md)"]
    Triage --> Artifact["Step 3: Render Audit Artifact<br/>(newsletter_subscription_audit_YYYY-MM-DD.md)"]
    Artifact --> Guidance["Step 4: Present Inline Summary &<br/>Gmail Search / Unsubscribe Checklist"]

Step 1 — Run Telemetry Script

Execute the bundled analysis script to obtain deterministic statistics across reports and suggestion files:

python3 .agents/skills/review-newsletter-subscriptions/scripts/analyze_subscriptions.py --days 30 --json

(Note: If the user specified a custom window like 60 days or explicit dates, pass --days N or --start-date YYYY-MM-DD --end-date YYYY-MM-DD.)

The script outputs structured JSON containing:

• reports_count: Total reports received per source in reports/Newsletter_*
• reviewed_count: Suggestions reviewed by user in data/suggestions_reviewed.md
• accepted_count / rejected_count: Feedback breakdown
• filtered_count: Suggestions blocked in data/suggestions_filtered.md with veto categories
• acceptance_rate ($A / R$) and yield_rate ($A / N$)
• Initial triage tag (Keep, Unsubscribe, Adjust, Watch)

Step 2 — Qualitative Triage & Contextual Analysis

Consult references/triage_criteria.md to evaluate each newsletter source against quantitative thresholds and qualitative project alignments:

1. 🟢 Keep (繼續訂閱):
   • Acceptance Rate $\ge 60%$ or Yield $\ge 20%$.
   • Has produced high-value project skills, prompt templates, or architecture improvements.
2. 🔴 Unsubscribe (建議取消訂閱):
   • Acceptance Rate $< 40%$ (with $\ge 3$ reviews), OR
   • High Hard-Veto ratio (e.g. Paper reading, Claude Code/Cursor), OR
   • High paywalled teaser / course sales marketing ratio, OR
   • High volume ($N \ge 10$) with negligible yield ($< 8%$).
3. 🟡 Adjust / Filter (設定 Gmail 篩選器):
   • High volume ($N \ge 15$), positive core value ($A \ge 2$), but contains separable noise (e.g. weekend specials, robotics).
   • Formulate targeted Gmail filter query (e.g. from:sender "Sunday Special").
4. ⚪ Watch (持續觀察):
   • Low volume ($N < 5$, $R < 2$) without strong negative signals, or suggestions currently in data/suggestions_pending.md.

Step 3 — Generate Audit Artifact

Create a dedicated Markdown Artifact using references/review_template.md:

• Target Path: <appDataDir>/brain/<conversation-id>/newsletter_subscription_audit_YYYY-MM-DD.md
• Title: # 📬 Newsletter Subscription Audit (YYYY-MM-DD ~ YYYY-MM-DD)
• Contents:
  1. Telemetry Table (Markdown table from analyze_subscriptions.py)
  2. 🟢 Keep: Highlight top articles, accepted suggestions, and workflow impact.
  3. 🔴 Unsubscribe: Detail why they failed (cite rejection comments and veto rules) + exact Gmail search queries.
  4. 🟡 Adjust & Filter: Specific Gmail filter rules to eliminate noise while preserving core signal.
  5. ⚪ Watch: Pending or low-sample notes.
  6. ✅ Action Checklist: Checkboxes with copyable Gmail search queries for manual execution.

Step 4 — Deliver Inline Summary in Chat

After rendering the artifact, present a concise executive summary directly in the chat conversation:

• Point the user to the generated audit artifact.
• Highlight the Top 3 to Keep and Top 3 to Unsubscribe with core rationale.
• Reiterate that destructive actions (unsubscribing or setting Gmail filters) remain strictly under human control, referencing the checklist in the artifact.