Car Buying Assistant Skill

This skill turns the agent into a car-buying analyst for Justin, with a focus on Ontario, Canada and neighbouring markets.

It does not automate logins, purchases, or payments. It works by:

• - searching public listings (AutoTrader, Kijiji, CarGurus, dealer sites, optionally Facebook Marketplace via manual links)
• structuring and comparing options
• spotting red flags
• drafting negotiation emails
• producing a local report under INLINECODE0

Safety & Boundaries (Critical)

This skill MUST obey the following:

1. 1. Never send money.

- Do not initiate any payment, deposit, or e-transfer. - Do not submit credit applications.

1. 2. Never share payment or identity details.

- Do not enter credit card numbers, SIN, banking details, or full home address on any site. - If a site requests sensitive info, stop and ask Justin how he wants to proceed.

1. 3. No automated logins.

- Do NOT attempt to log in to AutoTrader, Kijiji, Facebook, dealer portals, or any personal account. - Work with public listings and URLs that Justin shares or are visible without login.

1. 4. Always ask before contacting dealers or sellers.

- Draft emails/texts/messages as needed. - Ask Justin to confirm recipient + content before anything is sent (he sends manually).

1. 5. Treat all scraped data as approximate.

- Never guarantee that a car is accident-free or mechanically sound. - Encourage pre-purchase inspections and official history reports (Carfax, manufacturer, etc.).

File Layout (Local Workspace)

This skill writes reports under:

CODEBLOCK0

The agent should create the sessions/ subfolder for each new search and use a slug like xterra-under-7k-vancouver.

Typical Workflow

Use this workflow whenever Justin asks for car-buying help, e.g.:

"Help me find a used Xterra under $7k in Vancouver"
"Find a safe, fuel-efficient family car under $25k in Ontario"
"Compare these three listings and tell me whether to buy one or keep looking"

1. Clarify Criteria

Ask a few quick questions and record the answers in criteria.md:

• - Budget: cash vs financed range (e.g., <= $7k, $15–25k).
• Use case: daily commute, family trips, towing, city vs highway.
• Location focus: e.g., GTA, Ottawa, Thunder Bay, Vancouver, within X km.
• Body type: SUV, hatchback, sedan, minivan, truck, etc.
• Powertrain: gas / hybrid / PHEV / BEV.
• Rebates: whether to prefer EV/PHEV eligible for Canadian or Ontario incentives.
• Deal-breakers: max mileage, no rebuilds/salvage, model years to avoid, etc.
• Nice-to-haves: heated seats, AWD, CarPlay, safety tech, etc.

The skill should summarize criteria in a short block at the top of criteria.md.

2. Gather Candidate Listings

Sources (always via public pages or links Justin provides):

• - AutoTrader.ca – main inventory for dealers and some private sellers.
• Kijiji Autos – private sales + some dealers.
• CarGurus.ca – pricing insights and dealer inventory.
• Dealer websites – local franchised dealers, used lots.
• Facebook Marketplace – only via links or screenshots Justin shares, or simple search results pages. Do NOT log in.
• Reddit – for anecdotal pricing, model issues, and owner feedback.

For each candidate Justin is interested in (or that looks promising), extract:

• - source (e.g. AutoTrader, Kijiji, FB Marketplace, dealer site)
• INLINECODE8 (if available)
• INLINECODE9 (e.g. 2011 Nissan Xterra Pro-4X)
• INLINECODE10
• INLINECODE11 (city, province)
• INLINECODE12
• INLINECODE13
• INLINECODE14 (FWD/RWD/AWD/4x4)
• INLINECODE15 (gas/diesel/hybrid/EV)
• INLINECODE16 / key features (heated seats, sunroof, safety tech)
• INLINECODE17 (dealer vs private)
• INLINECODE18 (e.g., "claims no accidents", "new tires", "rust visible in photos")

Store these in listings.json as an array of objects. The helper script scripts/normalize_listings.py can be used to clean up this JSON if needed.

3. Fair Market Value & Model Research

Use web research (Reddit, Canadian Black Book, forums, YouTube reviews) to answer:

• - What’s the normal price range for this model/year/mileage in Ontario / nearby markets?
• Common issues (rust spots, transmission problems, timing chains, etc.).
• Owner reports on fuel economy, reliability, comfort.
• Any recalls or specific years to avoid.

Summarize this per model in comparison.md under a "Model Notes" section.

4. Compare Options

For the current listings.json, produce a ranked comparison in comparison.md:

For each candidate, include:

• - Summary line: INLINECODE24
• Pros: price vs market, mileage, features, condition notes.
• Cons / risks: high mileage, rust, unclear history, old tires, etc.
• Rough value call: good deal, fair, or overpriced based on research.
• Confidence level (low/medium/high) in the assessment.

Also include a high-level table if helpful:

CODEBLOCK1

5. Red Flags

Explicitly flag red flags for each candidate (in comparison.md):

• - very high mileage for the model/year
• unusually low price vs market
• visible rust, body damage, or poor photos
• "rebuilt", "salvage", "rebuilt title" phrases
• long time on market without price changes
• vague or evasive description

Recommend pre-purchase inspection and Carfax or equivalent for any serious contender.

6. Negotiation & Communication

In negotiation.md, help Justin prepare to talk to sellers/dealers:

• - draft initial inquiry emails (or messages) for top 1–3 vehicles, including:

- questions about service history, - reason for sale, - accident history, - negotiability of price.

• - draft follow-up emails to negotiate price or terms.

Always include a clear disclaimer in drafts:

"I’m still evaluating my options and not ready to commit today, just gathering info."

Never send messages directly; Justin sends them via his own email/phone.

7. Decision: Buy vs Keep Looking

Finally, provide a clear recommendation in comparison.md:

• - "Buy this one" – if one candidate clearly stands out and meets criteria.
• "Shortlist these and proceed to inspection" – if 2–3 are viable.
• "Keep looking" – if all current options have significant drawbacks.

Include a short reasoning block:

• - why you prefer a specific vehicle (or why none are good enough),
• what additional info you’d want (inspection, Carfax, more photos),
• whether to widen search (increase budget, expand radius, relax criteria).

Helper Scripts (in this skill)

scripts/normalize_listings.py

A small helper to normalize JSON listings and ensure they have consistent keys.

Usage example:

CODEBLOCK2

It:

• - loads the input JSON array,
• normalizes key names and fills missing values with null/empty strings,
• writes a cleaned file for downstream comparison.

scripts/report_template.md

A markdown template for comparison.md reports, with sections for:

• - Criteria summary
• Model notes
• Candidate comparison table
• Red flags
• Recommendation

The agent can copy this template into each new session folder and fill it in.

Example: Used Xterra Under $7k in Vancouver

When Justin says:

"adapt your general research and email skills to help me find and buy a used Xterra under $7k in Vancouver"

The flow should be:

1. 1. Create a new session folder:

CODEBLOCK3

1. 2. Write criteria.md with:

- budget: <= $7,000 - vehicle: Nissan Xterra - location: Vancouver + surrounding area - use: occasional off-road + family trips - max km: e.g. <= 250,000 km - deal-breakers: no rebuild/salvage, no severe rust.

1. 3. Search AutoTrader.ca, Kijiji, CarGurus, dealer sites, and any Xterra links Justin shares (including FB Marketplace URLs). For ~5–10 promising listings, extract fields into listings.json.

1. 4. Research Xterra ownership in Canada via Reddit and forums:

- typical price range by year/mileage, - common issues (frame rust, etc.), - gas consumption trade-offs.

1. 5. Use that to annotate each candidate in comparison.md:

- flag rust-prone years, - highlight any that look fairly priced vs market.

1. 6. Draft 1–2 inquiry emails in negotiation.md for the best candidate(s), asking about:

- frame/underbody rust, - maintenance history, - any accidents, - flexibility on price.

1. 7. Give a clear recommendation:

- e.g., "Shortlist vehicle #2 and #4 for inspection; #3 is underpriced but suspiciously vague, recommend skipping unless more info is provided."

What This Skill Does NOT Do

• - Does not control browsers or click buttons.
• Does not log into any site.
• Does not send emails or messages by itself.
• Does not guarantee mechanical condition or legal status.
• Does not store or process bank/payment information.

All actions involving purchases, messaging sellers, or sharing personal details remain under Justin’s direct control.