LeaseHackr Deal Report

Fetches lease-deal topics from LeaseHackr's public Discourse JSON API (no auth), filtered by vehicle class (e.g. "luxury suv") and location (e.g. "new jersey"), then renders a markdown report separating community-signed deals (real, reported by members) from broker/marketplace offers (paid advertisements). Signed deals are the priority throughout: they are actual transacted prices and the benchmark broker quotes should be judged against, so they lead the report, the takeaways, and the analysis effort. Offers that ship nationwide are included by default and marked as such — many brokers deliver anywhere.

Prerequisites

• curl and jq (check with command -v curl jq)
• minijinja-cli for report rendering (install: cargo install minijinja-cli)
• Network access to forum.leasehackr.com. No API key, no env vars.

Usage

scripts/fetch-deals.sh -c <class> -l <location> [options]

Output is one JSON doc: { query, signed: [...], marketplace: [...] }. Each topic has id, title, slug, url, brand, tags, created_at, bumped_at, posts_count, views, like_count, gist, availability, ships and (with --details) first_post_excerpt. availability is local (tagged with the requested location), nationwide (tagged usa-nationwide), or ships (carries the shipping tag). Topics tagged EXPIRED are dropped.

Workflow

1. Resolve inputs. Map the user's phrasing to a class and location key. Run scripts/fetch-deals.sh --list if unsure. "Luxury" classes filter to a luxury-brand set client-side (the forum's luxury tag is too sparse).
2. Fetch. For a report, always use details and hot: scripts/fetch-deals.sh -c "luxury suv" -l "new jersey" --days 30 --details 6 --hot 8 (use --details 0 --hot 0 only for a quick count/scan). Expect ~1s per API request; a full run takes 1–3 minutes.
3. Backfill signed deals if thin. Signed deals are the priority; a report with 0–4 of them is a weak benchmark. When the signed section comes back with fewer than 5 topics, run a second pass with a wider window and merge: scripts/fetch-deals.sh -c ... -l ... --category signed --days 90 --details 10 Keep the marketplace section at the original window (stale ads mislead; older signed deals are still useful reference points — note their age).
4. Extract deal terms per topic from title, gist, and first_post_excerpt: monthly payment ($N/mo), DAS/down/drive-off, term/mileage (36/10k), MSRP, money factor (MF .00xxx), residual %, % off MSRP, broker fee. Use — when a value is absent. Never invent numbers.
5. Score and rank by savings. For every deal with published numbers:
   • Effective monthly = (monthly × term + due-at-signing + broker/acq fees) ÷ term — folds DAS into the payment so a "$0 down" and a "$3k down" deal compare fairly. Exclude MSDs (refundable deposits).
   • Leasehackr Score — the community's native metric: the number of years of effective payments it would take to equal the sticker price. Official formula (per LeaseHackr's co-founder, https://forum.leasehackr.com/t/leasehackr-score-question/394887): Score = (MSRP × (1 + tax rate)) ÷ (Total Lease Cost ÷ months) ÷ 12 where Total Lease Cost = payment incl. tax × (months − 1) + drive-off
     • disposition fee − rebates − gov fees. In practice this reduces to Score = MSRP ÷ (12 × effective monthly) when payment and MSRP are both pre-tax. When the advertised payment includes tax, gross MSRP up by the stated tax rate first; if the rate isn't stated, skip the gross-up and label the score "conservative". Higher is better — observed range ~5–20 yrs; community rule of thumb: ~8+ solid, 10+ excellent, below ~6 weak (leasing that car is questionable).
   • Signed-deal posts usually link a Leasehackr Calculator page showing the official score — read it from the post when visible rather than recomputing.
   • If a deal's MSRP isn't posted but the identical model/trim/year has a published MSRP elsewhere in the same report, that may be used as a reference — prefix the score with ≈ and note the source. Rank each section by LH Score (highest first); deals with a payment but no computable score rank next by effective monthly; deals with no numbers or stale price signals (e.g. an excerpt dated a year ago) get rank — and sort last. Don't mix tax-inclusive and pre-tax numbers silently — note "+ tax" / "incl. tax" in the eff-monthly cell.
6. Spot limited & high-demand offers. Scan each marketplace topic's recent_posts (plus title/first_post_excerpt) for one-off specials with explicit scarcity or urgency language: "today only", "1 left" / "only N units", "will go quick / fast", "firesale", "blowout", "expires <date>", "last one", "price drop", "few remaining", "first come first served". Rules:
   • Require an explicit quantity or time limit — every ad says "great deal"; that alone does not qualify.
   • Only include posts from roughly the last 10 days (created_at on the post) — an expired flash deal is worse than none.
   • Quote the broker's own urgency phrase verbatim in the report; extract the vehicle and terms from the same post, — when absent.
7. Determine availability for each marketplace offer: start from the script's availability field, then refine from the excerpt — shipping language ("we ship", "delivery available", state coverage lists, delivery fee quotes) upgrades a nationwide offer to a confident "ships to <region>". Display strings like Local (NJ), Nationwide — ships, Ships to NJ (per post).

8. Assemble render data as JSON in a scratch file (arrays pre-sorted by rank). The vehicle and offer strings become the hyperlink text for url in the rendered tables — keep them concise and free of | and [] characters:

   {
     "class_display": "Luxury SUV", "location_display": "New Jersey",
     "days": 30, "generated_at": "2026-07-24", "has_results": true,
     "signed_days": 90,
     "signed": [ {"rank": 1, "vehicle": "...", "terms": "...",
                  "eff_monthly": "$276 incl. tax", "lh_score": "16.7 yrs",
                  "date": "YYYY-MM-DD", "url": "..."} ],
     "marketplace": [ {"rank": "—", "offer": "...", "signals": "...",
                       "eff_monthly": "—", "lh_score": "—",
                       "availability": "...", "updated": "YYYY-MM-DD",
                       "url": "..."} ],
     "hot_offers": [ {"offer": "...", "terms": "...", "lh_score": "10.3 yrs",
                      "urgency": "broker's verbatim scarcity phrase",
                      "posted": "YYYY-MM-DD", "availability": "...",
                      "url": "..."} ],
     "highlights": ["one-line market observations"]
   }

   Include signed_days only when the backfill widened the signed window (it switches the header to show both windows). Keep signed and marketplace strictly separate — real deals vs ads. highlights must lead with the best signed deal(s) as the benchmark, then the best broker offers measured against it. Omit hot_offers (or leave it empty) when nothing qualifies — the section only renders when present.

9. Render with the template next to this SKILL.md — do not hand-write the report: minijinja-cli template.j2 /path/to/data.json
10. Present, and offer to drill into a specific thread: curl -s -H 'User-Agent: lease-hackr-skill/1.0' https://forum.leasehackr.com/t/<id>.json | jq -r '.post_stream.posts[0].cooked'

Edge cases & notes

• Unknown class/location: the script exits 1 and prints the valid keys — relay the closest options to the user, or add a new entry to data/mappings.json for a genuinely missing region/class.
• Zero results: widen --days (60/90), fall back from luxury suv to suv, or retry with -l nationwide.
• Rate limiting: the script sleeps 1s between requests and backs off on HTTP 429 automatically. If a run still fails, wait a minute and retry — don't tighten the sleeps. Keep --pages/--details modest; this is a community forum, not an API product.
• Category IDs drift: signed = category 6, marketplace = 7 + regional subcategories 14–18. If sections come back empty despite raw matches, re-verify with curl -s https://forum.leasehackr.com/site.json | jq '.categories[] | {id, slug, parent_category_id}' and update data/mappings.json.
• Advertised prices usually exclude tax/fees and may require specific qualifications (loyalty/conquest, tier-1 credit) — surface this in Takeaways when the excerpts show it.
• This skill only reads public forum data — it never posts, messages, or authenticates.