| name | past-sales-agent |
| description | Use to log a completed used-machinery sale and to generate monthly / quarterly performance reports. Logs every detail of a blind co-broker deal — seller, third-party dealer, buyer, equipment (brand/model/year/serial/type), dimensions & weight, rigger, trucker, haul type, origin/destination, sale price, costs, gross margin — into the shared Sales tab of the dealer's Google Sheet. On request, generates a branded monthly PDF report aggregating by equipment type, party, geography, margin, and rigger/trucker mix. Triggers on "log this sale", "log the Krauss-Maffei sale", "month-end report", "monthly sales report", "Q1 numbers", "how many machines did I move", "what was my margin last month", "who was my best rigger this quarter", or any retrospective question about the dealer's book. NOT for prospective sales drafting (that's sales-agent). |
Past Sales Agent
The dealer's permanent record. Every completed deal gets logged with enough detail to reconstruct it months later — including the blind co-broker structure (who held it, who bought it, who was in the middle). Monthly reports aggregate the data into insights the dealer actually uses: best margins by equipment type, top riggers, freight cost trends, geographies that are hot.
When this fires
Logging
- "Log the Krauss-Maffei deal — sold to Midwest Molding, ABC was the seller, Bayshore rigged it"
- "Closed the Engel sale today, $52k, freight was $3,400"
- "Add to past sales: 2008 CX 160, blind, profit $8,200"
Reporting
- "Run the monthly report"
- "What did I sell in April?"
- "Show me Q1 numbers"
- "Which rigger ran the most loads last month?"
- "Best margin machine type this year?"
- "Generate the past-sales PDF for the board"
Setup
Uses the same Google Sheet as sales-agent. If .dealer_sales_sheet.json doesn't exist, point the dealer at sales-agent for bootstrap. Both skills read/write the same Sales tab.
Sales tab columns (full schema)
| Column | Notes |
|---|
sale_id | Auto: S-YYYYMMDD-#### |
sale_date | ISO date |
seller | Holder of the machine — the originating party (e.g. "ABC Equipment Inc.") |
third_party_dealer | The middle dealer if blind (the dealer running this skill, in most cases) |
buyer | End buyer |
blind_transaction | TRUE / FALSE — is this a blind co-broker deal? |
equipment_brand | Krauss-Maffei, Engel, Haitian, Arburg, etc. |
equipment_model | CX 160, Victory 200, Mars II, etc. |
equipment_type | injection_molding, cnc, presses, extrusion, blow_molding, robots, heavy_general |
year, serial | |
dimensions_lwh | e.g. 186" L × 64" W × 78" H |
weight_lbs | Integer |
rigger_company, rigger_id | rigger_id links to the rigging-agent database |
trucker_company, trucker_contact | |
haul_type | step_deck / RGN / flatbed / van / lowboy / expedited / multiple |
origin_city_state, destination_city_state | |
sale_price, cost_basis, rigging_cost, freight_cost | USD |
gross_margin | Auto-computed as sale_price - cost_basis - rigging_cost - freight_cost if not provided |
currency | Default USD |
proposal_number | Cross-ref to the sales-proposal PDF if applicable |
notes | |
Logging flow
- Gather facts — ideally pull from the relevant
sales-proposal PDF (if there's one with a matching proposal number) plus the dealer's verbal account. Confirm each detail before writing.
- Validate —
equipment_type must be enum, dimensions/weight should be numeric. Surface anything fuzzy and ask.
- Compute margin — if
gross_margin not provided, compute sale_price - cost_basis - rigging_cost - freight_cost. Surface it for sanity check before saving.
- Link the rigger — if the rigger isn't already in the rigging-agent database, prompt to add them. If they are, set
rigger_id and call rigging-agent's log_use() to update last-used-date and append history.
- Write the row —
scripts/log_sale.py appends to the Sales tab. Echo back the sale_id.
Reporting flow
scripts/monthly_report.py produces a branded PDF (matches the dark/gold aesthetic of the other dealer skills) covering:
- Header KPIs — # of sales, total revenue, total margin, average margin %, blind-deal % of book
- By equipment type — count, revenue, average margin per machine
- Top performers — best margin deal, fastest close, highest-revenue deal
- Logistics mix — top 3 riggers (count + total $), top 3 truckers, haul-type breakdown
- Geography heat — origin/destination state mix
- Trendline — vs. last month, vs. same month last year, YTD
- Parties — top 5 buyers, top 5 sellers (with blind-deal flag count)
- Watch list — anything margin-negative, anything over 90 days from sale-to-payment (if payment tracking is added later)
Output: /mnt/user-data/outputs/past_sales_<YYYY-MM>.pdf and a summary that goes in the conversation.
Quarterly and YTD reports use the same template with date-range parameter.
Anti-patterns
- ❌ Logging without confirming margin math — if the dealer says "margin was $8k" but the numbers don't add up, surface the discrepancy ("you said $8k but sale - cost - rig - frt = $7,150 — which is right?").
- ❌ Forgetting the
blind_transaction flag — this is the whole point of the dealer's book. Default to TRUE if third_party_dealer is set.
- ❌ Skipping
rigger_id linkage — losing the connection back to rigging-agent means the rigger's ratings/history degrade over time.
- ❌ Generating a report on a stale sheet — read fresh on every report run; don't cache.
- ❌ Reporting on a partial month without labeling it ("April so far" not "April final").
File map
scripts/log_sale.py — append a sale row + update related rigger recordscripts/monthly_report.py — generate the branded PDF + plain-text summaryscripts/report_metrics.py — pure-function aggregations (so the report can be embedded in other skills if needed)references/report_template.md — what each section of the PDF looks like
Companion skills
sales-agent — shares the same sheet (Customers + Leads + Outreach Log live there too)rigging-agent — sale logging calls log_use() to refresh the rigger's last-used infosales-proposal — links via proposal_number so you can jump from a sale row back to the executed PDF
