analyze-spending

name	analyze-spending
description	Analyze finances using LunchMoney API. Use when the user asks about spending patterns, budget analysis, subscription audit, recurring expenses, or financial overview for periods like "this month", "last month", "this year", or custom date ranges.

Analyze Spending with LunchMoney

Analyze your finances, identify spending patterns, review budgets, and find subscriptions you might want to cancel using data from LunchMoney.

Trigger Phrases

• /analyze-spending
• "analyze my spending"
• "review my finances"
• "find subscriptions to cancel"
• "audit my recurring expenses"
• "how much did I spend on [category]?"
• "budget review for [period]"
• "where is my money going?"

Prerequisites

1. LunchMoney Account: Active subscription with API access
2. API Key: Generate at https://my.lunchmoney.app/developers
3. 1Password Storage: Store the API key in 1Password:
   • Item name: Lunchmoney
   • Field: API Key

File Structure

skills/analyze-spending/
├── SKILL.md                      # This file
├── assets/
│   ├── MEMORY.md                 # Template for user preferences
│   └── subscription-audit-template.md  # Template for categorized report
└── data/                         # Runtime data (gitignored)
    ├── MEMORY.md                 # User preferences & decisions (copy from assets/)
    ├── cache/                    # Cached API responses
    │   ├── user.json             # User info (refreshed weekly)
    │   ├── categories.json       # Categories (refreshed weekly)
    │   ├── recurring_items.json  # Subscriptions (refreshed daily)
    │   └── transactions/         # Transaction data by month
    │       ├── 2025-12.json
    │       └── 2026-01.json
    ├── cache_meta.json           # Cache timestamps
    └── reports/                  # Generated markdown reports
        ├── 2026-01-19_spending-overview.md
        └── 2026-01-19_subscription-audit.md

---

Memory & Preferences

The MEMORY.md file stores user preferences, decisions, and learned patterns across sessions.

Initialize Memory

SKILL_DIR="skills/analyze-spending"
MEMORY_FILE="${SKILL_DIR}/data/MEMORY.md"
MEMORY_TEMPLATE="${SKILL_DIR}/assets/MEMORY.md"

# Create data directory and copy template if MEMORY.md doesn't exist
mkdir -p "${SKILL_DIR}/data"
if [ ! -f "$MEMORY_FILE" ]; then
  cp "$MEMORY_TEMPLATE" "$MEMORY_FILE"
  echo "Created MEMORY.md from template"
fi

Reading Preferences from Memory

Extract YAML configuration blocks from MEMORY.md:

# Extract essential subscriptions (won't flag for cancellation)
ESSENTIAL_SUBS=$(sed -n '/^essential_subscriptions:/,/^[a-z]/p' "$MEMORY_FILE" | \
  grep -E '^\s+-\s+payee:' | sed 's/.*payee:\s*"\(.*\)"/\1/')

# Extract excluded categories
EXCLUDED_CATS=$(sed -n '/^excluded_categories:/,/^[a-z]/p' "$MEMORY_FILE" | \
  grep -E '^\s+-\s+"' | sed 's/.*"\(.*\)"/\1/')

# Extract watchlist
WATCHLIST=$(sed -n '/^watchlist:/,/^[a-z]/p' "$MEMORY_FILE" | \
  grep -E '^\s+-\s+payee:' | sed 's/.*payee:\s*"\(.*\)"/\1/')

Updating Memory After Decisions

When the user makes a decision about a subscription, append to the Session History:

# Add a decision to memory
add_decision() {
  local date="$1"
  local subscription="$2"
  local decision="$3"
  local reason="$4"
  local savings="$5"

  cat >> "$MEMORY_FILE" << EOF

#### ${date}
- **${subscription}**: ${decision}
  - Reason: ${reason}
  - Annual savings: \$${savings}
EOF
}

# Example: User decided to cancel a subscription
add_decision "2026-01-19" "Crunch Gym" "cancel" "Keeping Chelsea Piers instead" "1053"

Adding to Watchlist

add_to_watchlist() {
  local payee="$1"
  local reason="$2"
  local review_date=$(date -v+30d +%Y-%m-%d)

  # This would need more sophisticated YAML editing
  # For now, prompt user to manually add:
  echo "Add to watchlist section in MEMORY.md:"
  echo "  - payee: \"${payee}\""
  echo "    added: $(date +%Y-%m-%d)"
  echo "    review_date: ${review_date}"
  echo "    reason: \"${reason}\""
}

Using Memory in Analysis

When analyzing, filter based on preferences:

# Filter out essential subscriptions from cancellation suggestions
filter_subscriptions() {
  jq --argjson essential "$(echo "$ESSENTIAL_SUBS" | jq -R . | jq -s .)" '
    map(select(.payee as $p | $essential | index($p) | not))
  '
}

# Exclude certain categories from spending totals
filter_transactions() {
  jq --argjson excluded "$(echo "$EXCLUDED_CATS" | jq -R . | jq -s .)" '
    .transactions | map(select(.category_name as $c | $excluded | index($c) | not))
  '
}

---

Workflow

0. Interview Mode (Always On)

The skill includes an interactive interview to validate subscription data and make decisions. This is critical because:

• LunchMoney's transactions_within_range field is often unreliable
• Users forget what's cancelled vs still active
• Categorization may need user confirmation

Interview Categories (ask about each in order):

1. Fitness - Gym memberships, fitness apps
2. Streaming - Video, music, audiobooks
3. AI/Dev Tools - AI subscriptions, hosting, developer tools
4. Reading/News - RSS, read-later, newsletters
5. Productivity - Software subscriptions
6. Memberships - Patreon, donations, co-ops

Interview Questions Pattern:

Use AskUserQuestion tool with:
- Show actual charges found from 12 months of transactions
- Group by category with totals (monthly/annual)
- Include "Last Charged" date and status indicators (✅/⚠️/🔴/❌)
- Distinguish cancelled (❌) from irregular (🔴 Xd ago)
- Offer actionable choices: Keep all, Cancel all, Review each, etc.

Key Learnings to Apply:

• Use 12-month window to catch annual subscriptions and show real activity
• Always cross-check recurring_items with actual transactions using normalized payee matching
• Show "Last Charged: Dec 2025" not confusing "Missed 3" status
• Price variations are normal (e.g., Starlink $120 → $5) - match by name not amount
• Ask follow-up questions based on responses
• Record all decisions in MEMORY.md

After Interview:

• Update MEMORY.md with decisions
• Generate action items list
• Calculate estimated savings
• Set review dates for watchlist items

1. Initialize Data Directory

SKILL_DIR="skills/analyze-spending"
DATA_DIR="${SKILL_DIR}/data"
CACHE_DIR="${DATA_DIR}/cache"
REPORTS_DIR="${DATA_DIR}/reports"
TRANSACTIONS_DIR="${CACHE_DIR}/transactions"

mkdir -p "${CACHE_DIR}" "${REPORTS_DIR}" "${TRANSACTIONS_DIR}"

# Initialize MEMORY.md if needed
MEMORY_FILE="${DATA_DIR}/MEMORY.md"
if [ ! -f "$MEMORY_FILE" ]; then
  cp "${SKILL_DIR}/assets/MEMORY.md" "$MEMORY_FILE"
fi

2. Retrieve API Key

LUNCHMONEY_API_KEY=$(op item get "Lunchmoney" --fields "API Key" --reveal)

3. Check Cache Validity

Before fetching, check if cached data is still fresh:

CACHE_META="${CACHE_DIR}/cache_meta.json"

# Initialize cache meta if it doesn't exist
if [ ! -f "$CACHE_META" ]; then
  echo '{}' > "$CACHE_META"
fi

# Check if cache is fresh (returns "true" or "false")
is_cache_fresh() {
  local cache_key="$1"
  local max_age_hours="$2"

  local last_fetch=$(jq -r ".${cache_key} // 0" "$CACHE_META")
  local now=$(date +%s)
  local age_hours=$(( (now - last_fetch) / 3600 ))

  [ "$age_hours" -lt "$max_age_hours" ] && echo "true" || echo "false"
}

# Update cache timestamp
update_cache_meta() {
  local cache_key="$1"
  local now=$(date +%s)
  local tmp=$(mktemp)
  jq ".${cache_key} = ${now}" "$CACHE_META" > "$tmp" && mv "$tmp" "$CACHE_META"
}

Cache freshness thresholds:

Data Type	Max Age	Rationale
user	168 hours (7 days)	Rarely changes
categories	168 hours (7 days)	Rarely changes
recurring_items	24 hours	May change daily
transactions (current month)	4 hours	Active updates
transactions (past months)	168 hours	Historical, stable

4. Fetch and Cache User Info

USER_CACHE="${CACHE_DIR}/user.json"

if [ "$(is_cache_fresh user 168)" = "false" ] || [ ! -f "$USER_CACHE" ]; then
  curl -s "https://dev.lunchmoney.app/v1/me" \
    -H "Authorization: Bearer ${LUNCHMONEY_API_KEY}" \
    > "$USER_CACHE"
  update_cache_meta "user"
  echo "Fetched fresh user data"
else
  echo "Using cached user data"
fi

5. Fetch and Cache Categories

CATEGORIES_CACHE="${CACHE_DIR}/categories.json"

if [ "$(is_cache_fresh categories 168)" = "false" ] || [ ! -f "$CATEGORIES_CACHE" ]; then
  curl -s "https://dev.lunchmoney.app/v1/categories" \
    -H "Authorization: Bearer ${LUNCHMONEY_API_KEY}" \
    > "$CATEGORIES_CACHE"
  update_cache_meta "categories"
  echo "Fetched fresh categories"
else
  echo "Using cached categories"
fi

6. Fetch and Cache Transactions

Transactions are cached per-month to enable efficient incremental updates:

fetch_transactions_for_month() {
  local year_month="$1"  # Format: YYYY-MM
  local start_date="${year_month}-01"
  local end_date=$(date -j -f "%Y-%m-%d" "${start_date}" -v+1m -v-1d +%Y-%m-%d 2>/dev/null || \
                   date -d "${start_date} +1 month -1 day" +%Y-%m-%d)

  local cache_file="${TRANSACTIONS_DIR}/${year_month}.json"
  local cache_key="transactions_${year_month//-/_}"

  # Current month: refresh every 4 hours; past months: refresh weekly
  local current_month=$(date +%Y-%m)
  local max_age=168
  [ "$year_month" = "$current_month" ] && max_age=4

  if [ "$(is_cache_fresh $cache_key $max_age)" = "false" ] || [ ! -f "$cache_file" ]; then
    curl -s "https://dev.lunchmoney.app/v1/transactions?start_date=${start_date}&end_date=${end_date}&debit_as_negative=true&limit=1000" \
      -H "Authorization: Bearer ${LUNCHMONEY_API_KEY}" \
      > "$cache_file"
    update_cache_meta "$cache_key"
    echo "Fetched transactions for ${year_month}"
  else
    echo "Using cached transactions for ${year_month}"
  fi
}

# Fetch for a date range (fetches all months in range)
START_DATE="2025-12-01"
END_DATE="2026-01-31"

# Generate list of months in range and fetch each
current="$START_DATE"
while [ "$current" \< "$END_DATE" ] || [ "$current" = "$END_DATE" ]; do
  month=$(echo "$current" | cut -d- -f1-2)
  fetch_transactions_for_month "$month"
  current=$(date -j -f "%Y-%m-%d" "${month}-01" -v+1m +%Y-%m-%d 2>/dev/null || \
            date -d "${month}-01 +1 month" +%Y-%m-%d)
done

7. Fetch and Cache Recurring Items

RECURRING_CACHE="${CACHE_DIR}/recurring_items.json"
START_DATE=$(date -v-60d +%Y-%m-%d)  # 60 days back for context
END_DATE=$(date +%Y-%m-%d)

if [ "$(is_cache_fresh recurring 24)" = "false" ] || [ ! -f "$RECURRING_CACHE" ]; then
  curl -s "https://dev.lunchmoney.app/v1/recurring_items?start_date=${START_DATE}&end_date=${END_DATE}&debit_as_negative=true" \
    -H "Authorization: Bearer ${LUNCHMONEY_API_KEY}" \
    > "$RECURRING_CACHE"
  update_cache_meta "recurring"
  echo "Fetched fresh recurring items"
else
  echo "Using cached recurring items"
fi

8. Load Cached Data for Analysis

# Read from cache files
USER_DATA=$(cat "$USER_CACHE")
CATEGORIES=$(cat "$CATEGORIES_CACHE")
RECURRING=$(cat "$RECURRING_CACHE")

# Merge transaction files for date range
TRANSACTIONS=$(jq -s '
  map(.transactions // []) |
  add |
  sort_by(.date) |
  reverse
' ${TRANSACTIONS_DIR}/*.json)

---

Report Generation

Generate Spending Overview Report

REPORT_DATE=$(date +%Y-%m-%d)
REPORT_FILE="${REPORTS_DIR}/${REPORT_DATE}_spending-overview.md"

generate_spending_report() {
  local start_date="$1"
  local end_date="$2"
  local period_name="$3"

  cat > "$REPORT_FILE" << EOF
# Spending Overview: ${period_name}

*Generated: $(date '+%Y-%m-%d %H:%M')*
*Period: ${start_date} to ${end_date}*

## Summary

| Metric | Amount |
|--------|--------|
| Income | \$${TOTAL_INCOME} |
| Expenses | \$${TOTAL_EXPENSES} |
| **Net** | **\$${NET}** |

## Spending by Category

| Category | Amount | % |
|----------|--------|---|
$(echo "$CATEGORY_BREAKDOWN")

## Top Merchants

| Merchant | Amount | # Txns |
|----------|--------|--------|
$(echo "$TOP_MERCHANTS")

---
*Data cached at: $(jq -r '.transactions_2026_01 // "N/A" | tonumber | strftime("%Y-%m-%d %H:%M")' "$CACHE_META" 2>/dev/null || echo "N/A")*
EOF

  echo "Report saved to: $REPORT_FILE"
}

Generate Subscription Audit Report (Template-Based)

The subscription audit uses a fill-in-the-blanks template at assets/subscription-audit-template.md. This approach:

• Categorizes subscriptions into meaningful groups
• Provides actionable recommendations at the top
• Analyzes 12 months of transaction data (default) for comprehensive coverage
• Cross-references recurring_items with actual transactions
• Shows "Last Charged" date instead of unreliable "Missed N" status
• Detects duplicate services and truly inactive subscriptions

Template placeholders: {{TOTAL_MONTHLY}}, {{CAT_STREAMING_ROWS}}, etc.

Subscription Categories:

Category	Keywords for Matching
🏠 Housing & Utilities	rent, house fee, loan, verizon, starlink
📺 Streaming	hulu, hbo, prime video, paramount, disney, spotify
💪 Fitness	gym, fitness, crunch
🧠 AI & Learning	chatgpt, jasper, execute program
💻 Developer Tools	github, linode, digitalocean, ngrok
📱 Productivity	obsidian, notion, adobe, zapier
📰 News & Media	substack, readwise, feedbin, newsblur
💾 Storage	backblaze, google storage, icloud
🎁 Memberships	patreon, rei, open source

Python-based generation (avoids jq != escaping issues in bash):

import json
from pathlib import Path

# Load data
with open("data/cache/recurring_items.json") as f:
    items = json.load(f)
with open("assets/subscription-audit-template.md") as f:
    template = f.read()

# Categorize and calculate
active_items = [i for i in items if not i.get("exclude_from_totals", False)]

def annual_cost(item):
    amount = float(item.get("amount", 0) or 0)
    cadence = item.get("cadence", "monthly")
    multipliers = {
        "yearly": 1, "monthly": 12, "weekly": 52,
        "biweekly": 26, "quarterly": 4
    }
    return amount * multipliers.get(cadence, 12)

# Generate recommendations based on:
# - Category overlap (multiple streaming, multiple gyms)
# - Inactive subscriptions (no transactions_within_range)
# - Duplicate payee names

# Fill template placeholders
output = template.replace("{{TOTAL_MONTHLY}}", f"{total:.2f}")
# ... fill all placeholders

Path("data/reports").mkdir(exist_ok=True)
with open(f"data/reports/{date}_subscription-audit.md", "w") as f:
    f.write(output)

Fetch 12 months of data for analysis (recommended default):

# Fetch 12 months of transactions for comprehensive analysis
# This catches annual subscriptions and provides accurate "Last Charged" dates
for i in {11..0}; do
  MONTH=$(date -v-${i}m +%Y-%m)
  fetch_transactions_for_month "$MONTH"
done

# Update recurring items with 90-day window (for LunchMoney's own tracking)
START_DATE=$(date -v-90d +%Y-%m-%d)
END_DATE=$(date +%Y-%m-%d)
curl -s "https://dev.lunchmoney.app/v1/recurring_items?start_date=${START_DATE}&end_date=${END_DATE}" \
  -H "Authorization: Bearer ${LUNCHMONEY_API_KEY}" \
  > "$RECURRING_CACHE"

Why 12 months?

• Catches annual subscriptions (GitHub, Backblaze, etc.)
• Shows true "Last Charged" date instead of "Missed N"
• Identifies cancelled vs just-irregular subscriptions
• Price changes are visible (e.g., Starlink $120 → $5)

---

Analysis Using Cached Data

Find ACTUALLY Active Subscriptions (Cross-Reference Method)

Important: Don't rely solely on recurring_items.transactions_within_range - it's often empty even when charges exist. Cross-reference with actual transactions using normalized payee matching:

import json
import re
from collections import defaultdict
from pathlib import Path

# Payee name normalization - critical for matching variations
def normalize(name):
    """Normalize payee name for matching (case-insensitive, strip suffixes)"""
    if not name:
        return ""
    name = name.lower().strip()
    # Remove common suffixes/prefixes
    name = re.sub(r'\s*(inc\.?|llc|ltd|co\.?|corp\.?|subscription|membership|payment|charge|premium)\s*', ' ', name)
    # Remove special characters
    name = re.sub(r'[^a-z0-9\s]', '', name)
    # Collapse whitespace
    name = re.sub(r'\s+', ' ', name).strip()
    return name

def get_keywords(name):
    """Extract brand keywords for fuzzy matching"""
    # E.g., "STARLINK INTERNET LLC" → ["starlink", "internet"]
    normalized = normalize(name)
    return set(normalized.split())

# Load 12 months of transactions from cache
transactions = []
for month_file in Path("data/cache/transactions").glob("*.json"):
    with open(month_file) as f:
        transactions.extend(json.load(f).get("transactions", []))

# Find transactions matching a recurring item (flexible matching)
def find_transactions_for_payee(target_payee, original_name=None):
    """Match transactions by normalized name or keywords"""
    target_norm = normalize(target_payee)
    target_keywords = get_keywords(target_payee)
    matches = []

    for txn in transactions:
        payee = txn.get("payee", "")
        payee_norm = normalize(payee)

        # 1. Direct normalized match
        if target_norm and target_norm in payee_norm:
            matches.append(txn)
            continue

        # 2. Reverse containment
        if payee_norm and payee_norm in target_norm:
            matches.append(txn)
            continue

        # 3. Keyword overlap (for brand names)
        payee_keywords = get_keywords(payee)
        if target_keywords & payee_keywords:  # Set intersection
            matches.append(txn)
            continue

    return matches

# Build subscription report with "Last Charged" dates
for item in recurring_items:
    payee = item.get("payee", "Unknown")
    txns = find_transactions_for_payee(payee, item.get("original_name"))

    if txns:
        # Sort by date descending
        txns.sort(key=lambda t: t.get("date", ""), reverse=True)
        last_charge = txns[0].get("date")  # e.g., "2025-12-22"
        charge_count = len(txns)
        monthly_avg = sum(abs(float(t.get("amount", 0))) for t in txns) / 12
        print(f"{payee}: ${monthly_avg:.2f}/mo | Last: {last_charge} | {charge_count} charges")
    else:
        print(f"{payee}: ❌ No charges found in 12 months (likely cancelled)")

Key matching strategies:

1. Name normalization: "STARLINK INTERNET LLC" → "starlink internet"
2. Keyword extraction: Matches "starlink" even if merchant name varies
3. Price-agnostic: Same vendor, different amounts still match
4. Bidirectional containment: "HBO Max" matches "HBO MAX SUBSCRIPTION"

Status indicators (in reports):

• ✅ Active: Last charged within 45 days
• ⚠️ Stale: Last charge 45-90 days ago
• 🔴 Xd ago: Shows days since last charge (e.g., "🔴 136d ago")
• ❌ Cancelled: No charges found in 12 months

Analyze Spending by Category

jq -r '
  .transactions |
  group_by(.category_name) |
  map({
    category: (.[0].category_name // "Uncategorized"),
    total: (map(.amount | tonumber | if . < 0 then . * -1 else 0 end) | add),
    count: length
  }) |
  sort_by(-.total) |
  .[:15]
' "${TRANSACTIONS_DIR}/2026-01.json"

Calculate Income vs Expenses

jq -r '
  .transactions |
  [
    (map(select(.amount | tonumber | . < 0) | .amount | tonumber | . * -1) | add),
    (map(select(.amount | tonumber | . > 0) | .amount | tonumber) | add)
  ] |
  {expenses: .[0], income: .[1], net: (.[1] - .[0])}
' "${TRANSACTIONS_DIR}/2026-01.json"

Find Top Merchants

jq -r '
  .transactions |
  map(select(.amount | tonumber | . < 0)) |
  group_by(.payee) |
  map({
    payee: .[0].payee,
    total: (map(.amount | tonumber | . * -1) | add),
    count: length
  }) |
  sort_by(-.total) |
  .[:10]
' "${TRANSACTIONS_DIR}/2026-01.json"

Subscription Annual Cost Analysis

jq '
  def annual_cost(amount; cadence):
    if cadence == "weekly" then amount * 52
    elif cadence == "biweekly" then amount * 26
    elif cadence == "twice a month" then amount * 24
    elif cadence == "monthly" then amount * 12
    elif cadence == "every 3 months" then amount * 4
    elif cadence == "every 4 months" then amount * 3
    elif cadence == "twice a year" then amount * 2
    elif cadence == "yearly" then amount
    else amount * 12
    end;

  map(select(.exclude_from_totals != true)) |
  {
    total_monthly: (map(if .cadence == "monthly" then (.amount | tonumber) else 0 end) | add),
    total_annual: (map(annual_cost((.amount | tonumber); .cadence)) | add),
    count: length
  }
' "$RECURRING_CACHE"

---

Force Refresh Cache

To force a fresh fetch of all data:

# Clear all cache
rm -rf "${CACHE_DIR}"/*
echo '{}' > "${CACHE_DIR}/cache_meta.json"
mkdir -p "${TRANSACTIONS_DIR}"

# Or clear specific cache
rm "${CACHE_DIR}/recurring_items.json"
jq 'del(.recurring)' "$CACHE_META" > tmp && mv tmp "$CACHE_META"

---

Viewing Past Reports

# List all generated reports
ls -la "${REPORTS_DIR}"

# View most recent spending report
cat "${REPORTS_DIR}/$(ls -t ${REPORTS_DIR}/*spending* | head -1)"

# View most recent subscription audit
cat "${REPORTS_DIR}/$(ls -t ${REPORTS_DIR}/*subscription* | head -1)"

---

Cadence Calculation

Convert LunchMoney cadence to annual cost:

Cadence	Multiplier
weekly	× 52
biweekly	× 26
twice a month	× 24
monthly	× 12
every 3 months	× 4
every 4 months	× 3
twice a year	× 2
yearly	× 1

---

Error Handling

Error	Cause	Solution
401 Unauthorized	Invalid/expired API key	Generate new key at lunchmoney.app/developers
404 Not Found	Invalid endpoint	Check API URL
Empty transactions	No data for period	Verify date range has transactions
1Password error	Item not found	Ensure "Lunchmoney" item exists with "API Key" field
Stale cache	Data not refreshing	Check cache_meta.json timestamps, force refresh if needed

---

API Reference

• Base URL: https://dev.lunchmoney.app/v1
• Authentication: Bearer token in Authorization header
• Documentation: https://lunchmoney.dev/
• Rate Limits: Standard API rate limits apply

---

Example Session

User: /analyze-spending

Claude: I'll analyze your spending. Let me check the cache and cross-reference
actual transactions (12-month window for comprehensive coverage)...

[Fetching 12 months of transactions: Feb 2025 - Jan 2026]
[Cross-referencing recurring_items with actual transaction data]
[Using normalized payee matching for accuracy]

Found 34 subscriptions with charges in the past year. Starting interview...

─────────────────────────────────────────────
📺 STREAMING & ENTERTAINMENT
   Active Monthly: $85.93 | Annual: $1,031.17

| Service      | Monthly | Last Charged | Status    |
|--------------|---------|--------------|-----------|
| HBO Max      | $15.82  | 2025-12-27   | ✅ Active |
| Paramount+   | $8.66   | 2025-12-24   | ✅ Active |
| Crunchyroll  | $5.33   | 2025-12-26   | ✅ Active |
| Hulu         | $3.16   | 2025-12-25   | ✅ Active |
| Audible      | $11.21  | 2026-01-04   | ✅ Active |
| Spotify      | $9.99   | Never (12mo) | ❌ Cancelled |
| Disney+      | $7.99   | Never (12mo) | ❌ Cancelled |

Which do you want to keep?
[ ] Keep 2-3 favorites
[ ] Rotate quarterly
[x] Cancel all

User: Cancel all
─────────────────────────────────────────────

─────────────────────────────────────────────
💻 DEVELOPER TOOLS & HOSTING
   Active Monthly: $48.54 | Annual: $582.43

| Service             | Monthly | Last Charged | Status      |
|---------------------|---------|--------------|-------------|
| GitHub              | $15.33  | 2025-09-05   | 🔴 136d ago |
| Google Cloud Storage| $14.61  | 2026-01-07   | ✅ Active   |
| Ngrok               | $10.50  | 2026-01-19   | ✅ Active   |
| DigitalOcean        | $8.10   | 2025-12-31   | ✅ Active   |

User: Cancel DO and Ngrok
─────────────────────────────────────────────

[Continues through all categories...]

## Summary of Decisions

| Action | Services | Monthly Savings |
|--------|----------|-----------------|
| Cancel | Streaming (5 active) | $44 |
| Cancel | Dev tools (2 services) | $19 |
| Cancel | Productivity (4 services) | $80 |
| Downgrade | Claude.ai ($218→$100) | $118 |
| **Total** | | **$261/mo ($3,132/yr)** |

Action items saved to: data/MEMORY.md
Report saved to: data/reports/2026-01-19_subscription-audit-12mo.md

Next steps:
- [ ] Cancel streaming services (HBO, Paramount+, Crunchyroll, Hulu, Audible)
- [ ] Cancel DigitalOcean and Ngrok
- [ ] Downgrade Claude.ai to $100 tier
- [ ] Review Patreon creators by 2026-02-01

---

Notes

• All amounts are in the user's primary currency unless specified
• Debit transactions should be treated as negative (expenses)
• Credit transactions are positive (income/refunds)
• Don't trust recurring_items.transactions_within_range - always cross-reference with actual transactions
• Default analysis window is 12 months to catch annual subscriptions and show accurate "Last Charged" dates
• The API returns a maximum of 1000 transactions per request; use pagination for larger date ranges
• Cache is stored in data/ which should be gitignored
• Reports are timestamped for historical comparison

Status Legend (in reports)

Status	Meaning
✅ Active	Last charged within 45 days
⚠️ Stale	Last charge 45-90 days ago (may be cancelled or annual)
🔴 Xd ago	Shows days since last charge (e.g., "🔴 136d ago" for GitHub)
❌ Cancelled	No charges found in 12 months

Payee Matching Notes

• Names are normalized: "STARLINK INTERNET LLC" → "starlink internet"
• Matching is price-agnostic (same vendor, different amounts still match)
• Keyword extraction catches brand name variations
• Apple charges are mixed (TV, iCloud, App Store) - hard to separate by merchant name alone