subscription-audit
GitHub用于审计订阅支出，通过系统化查找隐藏扣费、年化排名及评估使用频率，提供保留/取消/降级建议。支持脚本自动化计算年度成本，帮助用户识别并削减不必要的经常性开支。
Trigger Scenarios
Install
npx skills add mohitagw15856/pm-claude-skills --skill subscription-audit -g -y
SKILL.md
Frontmatter
{
"name": "subscription-audit",
"description": "Find and rank the recurring-payment leak — every subscription annualized, sorted by real yearly cost, with the keep\/cancel\/downgrade pass and the where-they-hide checklist. Use when asked audit my subscriptions, how much am I spending on subscriptions, help me cancel stuff, or what recurring charges am I forgetting. Produces the annualized ranking from the script, the hidden-subscription hunt list, the keep\/cancel\/downgrade decisions with the cancellation friction notes, and the re-audit cadence."
}
Subscription Audit Skill
Subscriptions are priced monthly precisely so nobody computes them yearly — $15.99 is a sandwich; $191.88 is a decision. The audit is two passes: find them all (the hunt is the hard part — they hide across cards, app stores, PayPal, and annual charges that only surface one month a year) and annualize and rank them, at which point most of the leak turns out to live in the top three lines. This skill runs both passes and then the honest keep/cancel/downgrade sort — including the note that canceling is sometimes deliberately harder than subscribing, and how to do it anyway.
What This Skill Produces
- The annualized ranking — every recurring charge at its true yearly cost, from the script, with per-month and per-day totals
- The hunt checklist — the places subscriptions hide, worked through systematically
- The sort — keep (used, valued) / cancel (the honest list) / downgrade (the tier nobody remembers choosing) / negotiate (the ones that discount on cancel-intent)
- The re-audit cadence — because the leak refills
Required Inputs
Ask for these if not provided:
- The recurring lines — from bank/card statements (2–3 months back, plus one full year scan for annual charges); raw pasted statements are fine — extraction is part of the job
- The hunt surfaces — which cards, app-store subscriptions (both platforms), PayPal/payment-app recurring, anything on a partner's card that's really shared
- Honest usage — per service: when last actually used (the calendar answer, not the aspirational one — "I might get back into it" is the leak talking)
Programmatic Helper
python3 scripts/subscription_audit.py --sub "streaming-a:15.99:monthly" --sub "vpn:71.88:yearly" --sub "gym:45:monthly"
python3 scripts/subscription_audit.py --sub "news:4:weekly" --sub "cloud:2.99:monthly" --json
Deterministic. Cadences: weekly/monthly/quarterly/yearly; output ranks by annualized cost and reports the top-three share — usually most of the leak.
Framework: The Audit Rules
- The hunt is systematic or it's partial: card statements (all cards) → app-store subscription pages (both ecosystems — the ones subscribed via app store don't show as merchant names you recognize) → PayPal/payment-app recurring lists → the annual-charge sweep (scan 12 months, not 2 — domains, memberships, and software renew yearly and hide 11 months at a time) → the free-trial calendar (what converts next month?).
- Annualize before judging: every decision is made against the yearly number; the script exists because $2.99 and $45 monthly feel the same and are not. The per-day total is the motivation number; the ranking is the action list.
- The sort has four bins, not two: keep (used this month, priced fair) · cancel (the calendar says so) · downgrade (the forgotten premium tier — often the biggest recovery per minute of effort) · negotiate (services with retention offers: starting the cancel flow legitimately surfaces the discount; taking it sets a calendar reminder for when it expires).
- Duplicates and bundles get a pass of their own: two cloud storages, three streamers rotated seasonally (the rotation is the strategy: subscribe the month you watch, cancel after), a bundle that quietly covers a standalone being paid separately.
- Cancellation friction is real and beatable: note per-cancel what it takes (button / chat / phone call — the phone-call ones are that way on purpose); calendar the annual renewals two weeks ahead of their date, because the re-decision beats the auto-renew. Re-audit every 6 months — the leak refills at roughly one new subscription a month.
Output Format
Subscription Audit: [N] found — [total]/year
The Ranking
[Script output: annualized table, per-month/per-day, top-three share]
The Hunt Ledger
[Surfaces checked ✓ · surfaces still to check · trials converting soon]
The Sort
| Subscription | /year | Last used | Decision | Action + friction note |
|---|
Recovered: [the cancel+downgrade total]/year · Renewal calendar: [annual charges, dated two weeks early]
Educational model, not financial advice — and the rotation move (subscribe when using, cancel after) is allowed to be the whole strategy.
Quality Checks
- The hunt covered app stores and payment apps, not just card statements
- The annual-charge sweep scanned a full 12 months
- Every decision was made against the annualized number
- Downgrade and negotiate bins were considered, not just keep/cancel
- The recovered total and the renewal calendar both appear
Anti-Patterns
- Do not audit only the obvious card — the app-store and PayPal surfaces are where they hide
- Do not judge at monthly prices — annualize first, always
- Do not let "might use it again" survive contact with the last-used date — the calendar votes, aspiration doesn't
- Do not cancel the negotiables without walking the retention flow once — the discount is sitting right there
- Do not shame the keeps — a used, valued subscription is fine; the audit hunts the forgotten, not the enjoyed
Version History
- 54fad50 Current 2026-07-19 13:44
