AI Skills for Claude, ChatGPT, Codex and Cursor

Before: Every Friday I match last week's card charges to receipts and chase missing ones, 50 min. Receipts in Gmail, card CSV in Drive. Write to my Expenses 2026 sheet. Never email vendors without me, never touch past months. Friday 4pm, Lisbon.

After: Ship, 16/16. Expense Manager (Grok Bot), every Friday 4:00 PM Europe/Lisbon: matches charges to receipts in Expenses 2026, keyed by Transaction ID, and drafts one chase email per merchant. Nothing is sent. No CSV modified in the last 7 days means ROUTINE FAILED, never last week's file. Time saved: 40 min/week. Comes with the profile, 3 skills, 2 routines and a test run on a copy of the sheet.

Before: Lena Zhou's 1,400-word Fernway blog post: 7 onboarding emails on a timer got 9% clicks. Making them shorter changed nothing. 11 onboarding calls showed two stuck points. 4 emails sent when users stall got 31%.

After: I made our onboarding emails shorter. Click rate didn't move. Then I listened back to 11 setup calls. Nobody mentioned the emails. They mentioned two screens where they got stuck. I'd been editing the wrong thing. X post 3 of 4 (Confession, 215 chars). The same pack adds a 7-post thread, a LinkedIn post, a newsletter section, a video script and a RED note: 做SaaS的，新手引导邮件我只留了4封

Before: Our hero, built in v0: violet-to-blue gradient, centered H1 "Supercharge Your Workflow", subhead #94A3B8 on the gradient, two pill buttons, six emoji feature cards below. Does this look AI-made?

After: Score: 15/35 Verdict: Rework Generic tells found: violet gradient, everything centered, card soup, emoji icons, hype copy. 1. Accessibility: subhead #94A3B8 on #7C3AED is 2.22:1. Fix: solid #FBFAF7 hero, subhead #5A635E (5.95:1). Priority 9 2. Hierarchy: two equal pill buttons. Fix: one bg-primary button, the other a text link. Priority 9 3. Color: drop the gradient; one flat primary, for actions only. Priority 6

Before: Branch feat/ENG-482-webhook-retries. Commits: "wip retries", "fix tests". Diff adds retry columns and CREATE INDEX on webhook_deliveries, 30 s doubling backoff up to 8 attempts, and a worker behind WEBHOOK_RETRY_ENABLED. No CI run shared.

After: Summary: Failed webhook deliveries are now retried, up to 8 attempts with delays doubling from 30 seconds (about 64 minutes), behind WEBHOOK_RETRY_ENABLED, off by default. Risk: High. New index on webhook_deliveries (size unknown), and delivery becomes at-least-once. Rollback: Set WEBHOOK_RETRY_ENABLED=false. Safe. Must review 1. [Tier 1] retry-worker.ts: FOR UPDATE SKIP LOCKED runs outside a transaction, so two runs can send the same delivery twice.

Before: Card ending 4417, Jul to Sep: ACME FITNESS 49.00 monthly (twice on 09/12). ORBIT+ 0.00 on 07/18, then 15.99 monthly. STREAMBOX 15.49, then 17.99 in Sep. TIDEWAVE MUSIC 11.99. PAYPAL *LUMENMUSIC 10.99. Last gym visit: June.

After: Acme Fitness, $49.00 a month: cancel candidate. No visits since June, and it charged you twice on Sep 12, so ask for $49.00 back. Orbit+, $15.99 a month: cancel candidate. A $0 trial on Jul 18 that converted on Aug 17. Streambox: review. Up 16% in September, to $17.99. Tidewave Music and Lumen Music: ask household. Two music services, $22.98 a month. Potential savings: $64.99 a month, $779.88 a year, plus the $49.00 refund.

Before: Wed Oct 7, Zoom transcript: [02:30] Priya: ok so let's do it. Beta goes to those 20 accounts on Tuesday. [04:40] Marco: I think we're all aligned on killing the CSV import, right? [04:44] Dana: yeah makes sense [04:46] Speaker 3: hmm, Acme uses it every Monday for their roster upload

After: D1. Onboarding v2 beta goes to the 20 opted-in accounts on Tue Oct 13. Decided by: Priya Raman. A1. Priya Raman: Ask Acme whether they still need CSV import for their Monday roster upload. Due: Fri Oct 9 [suggested]. Q1. Do we remove CSV import in v2? Owner: Priya Raman. Needed by: Mon Oct 12. FLAGS - Fake decision: "aligned on killing the CSV import" [04:40] was contested at [04:46]. Logged as Q1, not a decision.

Before: Audit fernway.example/blog/reduce-patient-no-shows. It used to be our best post. Now it's on page 2, and ChatGPT and Perplexity cite Orbit Practice and a Reddit thread, never us.

After: Blocker: robots.txt disallows OAI-SearchBot and PerplexityBot site-wide, so ChatGPT search and Perplexity cannot cite any Fernway page. Fix: give them their own Allow group and keep GPTBot blocked if training is the concern. Next, High: the answer sits 430 words down. Open with: "To reduce no-shows, send two reminders (48 hours and 2 hours before), take a card on file at booking and offer a one-tap reschedule link." 12 more fixes, ranked.

Before: Acme went quiet. Proposal sent Tue Oct 6 to Lena Ortiz: 40 seats, $28,800/yr. On the demo, Raj (their security lead) said nothing goes live without a SOC 2 report. We can share ours under NDA. Today is Mon Oct 12. Write the nudge.

After: Subject: reply in thread (Acme proposal: 40 seats, annual) Hi Lena, Raj said on the demo that nothing goes live without a SOC 2 report. Ours is ready to share under NDA, so his review can start now instead of after you sign. Should I send the NDA to Raj directly, or would you rather loop him in? Jordan