buy-me-a-car

16 Claude Code skills that turn a used-car weekend into a 2-hour decision: scrape 9 sites in parallel, draft counter-offers, print a buyer-grade dossier.

English | 中文版

⭐ Read this first, the design philosophy

A dealership wins because it controls the frame: it splits the deal into sale price, tax, fees, and financing so each looks small, refreshes its inbox on its own schedule, and counts on you walking in cold. This plugin inverts every one of those advantages. The whole design follows from five iron rules, each born from a specific dollar-loss incident:

1. OTD only, never negotiate sale price / tax / fee separately, only the total out-the-door number. Splitting the deal is how you lose track of $1k at a time.
2. Plain ASCII emails, no markdown, em-dash, or smart quotes. Dealer mail clients render them as literal garbage characters, and a sloppy email reads like a sloppy buyer.
3. 3-anchor counters, every reply cites (a) the dealer's own internal price spread, (b) a regional market comp, and (c) your locked competitor OTD. Three anchors leave nothing to argue with.
4. 15-min cron sweep, Gmail is polled automatically; a human never manually refreshes. The buyer who replies first holds the leverage.
5. Walk-away threshold, over budget or counter rejected = a polite walk. Preserving optionality across many dealers beats grabbing one mediocre deal.

Everything downstream, the parallel scrape, the OTD calculator, the dossier, exists to make those five rules cheap to follow. The full rule set lives in SKILL.md.

What it is (and isn't)

It is a buyer-side negotiation cockpit: 1 orchestrator that runs a 9-phase pipeline (research → outreach → negotiate → close) plus 15 narrow sub-skills you can call standalone for single tasks (OTD math, state-fee lookup, CARFAX review, lease-vs-cash, trade-in valuation, and more). It carries real data, 50 states + DC of fee detail, 16-brand CPO eligibility, 6 buyer paths including private-party.

It isn't a price-prediction oracle, a dealer-side CRM, or tax / legal / financial advice. It does not auto-send anything binding, emails are saved as Gmail drafts for you to review and send. It is single-author alpha (see Limitations).

Install

/plugin install github:DaizeDong/buy-me-a-car

Or clone manually:

git clone https://github.com/DaizeDong/buy-me-a-car.git ~/.claude/plugins/buy-me-a-car

Skills auto-activate on phrases like help me buy a car, draft a reply to dealer, compute OTD, review CARFAX, etc.

Verify the install:

cd ~/.claude/plugins/buy-me-a-car
ls skills/   # should show 16 directories
python skills/orchestrator/scripts/otd_calculator.py --state NJ --sale-price 25000
python skills/orchestrator/scripts/generate_dossier.py \
  --config skills/orchestrator/assets/dossier_config_template.yaml \
  --output /tmp/test.html

Any failure → open an issue with the traceback. Most failures are Python deps (PyYAML, Jinja2) or Chrome headless path.

60-second tour

You say:

help me buy a used compact SUV this week, budget $25k OTD, under 60k miles, near <ZIP>

What runs automatically:

1. Confirms 9 criteria + buyer-type router (cash / financing / trade-in / EV / pickup / private-party)
2. Parallel subagents scrape 9 sites (Carfax, CarMax, Carvana, Cars.com, AutoTrader, Edmunds, TrueCar, CarGurus, Enterprise) and dedupe by VIN
3. Submits lead forms to top 30 candidates via Playwright MCP (anti-bot aware)
4. 15-min cron monitors Gmail; triages dealer replies into 4 buckets (real / OOO / CRM / spam)
5. Drafts plain-ASCII counter-offers using 3-anchor logic (dealer's own price spread + regional comp + your locked OTDs)
6. Extracts red flags from dealer-attached CARFAX / proposal PDFs
7. Generates 8-page dossier (HTML → headless-Chrome PDF, EN or CN template)
8. Close-day checklist with verbatim F&I add-on hard-no script

Output: typically $5-9k saved vs walking in cold.

Skills at a glance

16 skills total: 1 broad orchestrator + 15 narrow-trigger sub-skills. Sub-skills work standalone; the orchestrator routes to them inside the 9-phase pipeline.

How to invoke each skill

Each block: when to use, trigger phrases, one-line example, what comes back.

orchestrator

• Use when: you want the full 9-phase pipeline from scratch.
• Triggers (the complete set; the frontmatter description carries the first 12, which is all a 180-char cap holds):
  • EN: buy me a car, find me a car, email dealers, negotiate OTD
  • ZH: 帮我找车, 买车, 选车, 砍价, 对比经销商
  • ES: ayudame a comprar un carro, ayudame a comprar un coche, encuentrame un auto
  • Also fire the skill, but live only here: research cars, compare quotes, buy a Subaru/Toyota/Honda/Mazda, negociar el precio del carro, comparar concesionarios
• Example: help me buy a 2022-2024 Outback Premium under 60k miles within 50mi of <ZIP>, budget $32k OTD
• Output: phase-by-phase artifacts under a car_buying_<YEAR>/ working dir.
• 6 buyer paths: cash / financing / trade-in / EV / pickup, plus private-party, seller is a private individual (FSBO), not a dealer. No OTD stack and no F&I office; buyer pays tax + title + registration at the DMV. The work shifts to title transfer, state tax basis (purchase price vs book value vs Illinois-style fixed table), curbstoner detection, and payment/escrow safety (cashier's check at the seller's bank; pay any lien off directly to the lienholder). See skills/orchestrator/references/private_party_playbook.md.

otd-calculator

• Use when: convert sale price → OTD, or reverse-engineer max sale from target OTD.
• Triggers: compute OTD, OTD math, 算 OTD, 算总价
• Example: compute OTD for $30k sale in NJ with $499 doc fee
• Output: itemized OTD (tax / doc / title / reg / DMV) for all 50 states + DC (every state at fee-detail depth; 34 web-verified).

state-fee-lookup

• Use when: pull the 6-field summary (rate / local / doc cap / title / reg / trade credit) for any state.
• Triggers: doc fee in NJ, state 税率, trade-in credit
• Example: what's TX doc fee cap and EV reg surcharge
• Output: state summary + "Does NOT have" leak-detection list (catches NJ tire fee on a CT quote, etc.). Backing data: 50 states + DC, every state at fee-detail depth (30 full + 21 stub), 34 web-verified.

cpo-eligibility

• Use when: verify factory CPO eligibility + embedded $ value before paying the CPO premium.
• Triggers: is this car CPO, Subaru CPO, Honda Certified, CPO 资格
• Example: check CPO on 2021 Kia Telluride @ 55k miles
• Output: eligibility verdict (16-brand matrix across 12 programs: 8 mainstream + Stellantis SPOTiCAR [Ram/Jeep/Chrysler/Dodge/Fiat, 5 brands in one program] + luxury Lexus/Genesis/Acura), embedded $1-3k value, fake-CPO red flags.

carfax-pdf-review

• Use when: dealer sent you CARFAX / service-record / F&I-proposal PDFs.
• Triggers: review this CARFAX, 审 PDF, F&I add-on detection
• Example: review the CARFAX dealer just emailed
• Output: structured red-flag report (accidents, service gaps with $ ranges, 12 challengeable F&I add-ons).

dealer-reply-drafter

• Use when: draft ONE outbound reply, counter / follow-up / walk-away.
• Triggers: draft counter to dealer, 回复 dealer, 对 dealer 报价做 counter
• Example: draft a counter to this Honda dealer's $33k OTD, target $30.75k
• Output: Gmail draft (saved, not sent), ~10 lines, plain ASCII, 3 asks + 1 anchor + 1 walk-away.

inbox-triage

• Use when: dealer inbox piling up, separate real replies from CRM noise.
• Triggers: check my dealer inbox, 看下邮箱, dealer 回复了吗
• Example: triage today's dealer inbox
• Output: per-bucket counts (real / OOO / CRM / spam), Gmail labels applied, handoff list to dealer-reply-drafter.

dossier-builder

• Use when: print a buyer-grade research packet before the test drive.
• Triggers: build dossier, 生成 PDF, make dossier
• Example: generate the dossier PDF in Chinese template
• Output: 8-page HTML + headless-Chrome PDF (EN or CN), covering market avg / OTD / CPO embedded value / dealer anchor analysis.

ev-buyer-helper

• Use when: buyer is going EV, state/local rebate stack, charging (NACS/CCS1), used-EV battery diligence. NOTE: the federal §30D ($7,500 new) / §25E ($4,000 used) / §45W (lease pass-through) credits were TERMINATED for vehicles acquired after 2025-09-30 (OBBBA, Public Law 119-21); they are historical only and not counted in net-price math.
• Triggers: EV federal credit, $7,500 POS, 电车补贴
• Example: what EV rebates still apply in NJ for an Ioniq 5 SEL in 2026
• Output: net price after any state/local rebate (no federal credit) + NACS/CCS1 adapter guidance + used-EV SoH diligence.

payment-method-decider

• Use when: choose close-day instrument, cashier's check / credit card / wire / lease cap reduction.
• Triggers: cash or CC for car, 支付方式, Visa for $30k car
• Example: should I put $30k on my 3% cashback Visa or cashier's check
• Output: method recommendation with CC-rewards-vs-surcharge break-even math.

lease-vs-cash-analyzer

• Use when: dealer offered a lease, verify MF / residual / acquisition / disposition.
• Triggers: lease or buy, money factor markup, 租还是买
• Example: is this Ioniq 5 SEL lease quote at $575/mo honest
• Output: monthly breakdown + LEASE / BUY / BREAK-EVEN verdict by ownership horizon.

trade-in-valuator

• Use when: trading in, 4-anchor valuation + lien payoff workflow.
• Triggers: valuate my trade, 评估置换车, trade-in tax credit
• Example: what's my 2017 Civic worth as trade in NJ
• Output: 4-anchor table (KBB Instant / Trade-in / Private / Wholesale) + TRADE vs SEPARATE-SELL decision.

quote-evidence-collector

• Use when: collect REAL dealer-quote screenshots from XHS / Reddit / FB as negotiation anchors.
• Triggers: find quote screenshots, 搜集报价截图, find dealer evidence
• Example: find XHS quotes for 2024 Outback Premium in <your state>
• Output: REAL-tagged compressed _FINAL_*.jpg (1300px, 100-300 KB) ready for manual paperclip.

insurance-shopper

• Use when: setting up auto insurance before close day, new driver, cash buyer, or cross-state move.
• Triggers: set up insurance, car insurance quote, new driver insurance, 上保, 保险报价
• Example: set up insurance for a new SUV in <your state>, first-time driver
• Output: 3-carrier quote comparison (NJM / Geico / Progressive), 6-month total, recommended coverage spec, bind sequence.

ppi-scheduler

• Use when: ready to book pre-purchase inspection, mobile-PPI services by region.
• Triggers: book PPI, 提车前检车, mobile inspector
• Example: book a mobile PPI tomorrow for a 2022 Outback at a local dealer
• Output: bookings (ID + cancel deadline) + post-inspection PROCEED / COUNTER / WALK matrix.

close-day-checklist

• Use when: tomorrow is close day, buyer-type checklists + F&I add-on hard-no script.
• Triggers: ready to close, F&I add-on refusal, 提车清单
• Example: give me the close-day checklist for tomorrow cash buyer with trade-in
• Output: pre / on-site / post checklists + verbatim F&I scripts.

Trigger routing

When a query could activate multiple skills, the most narrow + specific trigger wins:

If ambiguous, name the skill explicitly: use dealer-reply-drafter to draft this. Unsure which one? Call orchestrator, it routes internally.

Example output

Phase 3 renders two Markdown tables. A site-capability matrix (which sources
yielded usable inventory/pricing, ranked into tiers):

...and a VIN-deduped candidate list with deal tags (all values below are
synthetic illustrations, not a real search):

Limitations

• Single-author alpha, workflow based on a single purchase cycle + 8 worked example scenarios (see examples/). Not multi-market validated.
• Data drifts, state fees, CPO terms, and EV credits were last verified 2026-05-18; major bills may have passed since.
• Anti-bot fragile, CarGurus / Cars.com / AutoTrader / Edmunds / TrueCar depend on Playwright MCP and may break with site redesigns.
• Not tax / legal / financial advice, verify with licensed professionals before signing.

Languages

Three languages, two surfaces, keep them separate:

• Buyer-facing chat + criteria.md + dossier: English, 中文, or Español. Triggers fire in all three (buy me a car / 帮我买车 / ayudame a comprar un carro). The dossier ships an EN and a CN print template; Spanish support is buyer-facing chat + an explanatory ES glossary for load-bearing terms (OTD, doc fee, ADM, CPO, NACS, GAP, MSRP), with regional carro / coche / auto mirroring.
• Dealer-facing email: always English + plain ASCII, regardless of the buyer's chat language. Dealer CRM clients mangle non-ASCII, and an English OTD ask threads cleanly with the dealer's own quote. The buyer reads the deal in their language; the dealer reads the ask in English. See Language and Audience Separation in skills/orchestrator/SKILL.md.

ES glosses are working translations pending native-speaker sign-off before production use.

This repo ships docs in two languages: English (README.md, authoritative) · 中文 (README_CN.md).

Roadmap · Contributing · License

ROADMAP.md tracks v0.3.0 / v1.0.0 plans (multi-author data, adversarial dealer tests, EV-credit reinstatement watch, remaining luxury-brand CPO). Pick one → open issue → PR. Changes are logged in CHANGELOG.md.

MIT, fork it, ship it, save someone money. See LICENSE.

last_verified: 2026-05-18