Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Scan configured sources (Craigslist now; eBay + Facebook Marketplace planned) for used-item listings matching the owner's criteria. Currently configured for a Mac mini search (M2+, 16GB+, 512GB+, ≤$500, near 94566). Notify owner via SMS + Telegram on a match.
.claude/skills/sonichi-deal-finder/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 167% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 76% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 6% | 0% |
Watches configured sources for second-hand item listings matching owner-defined criteria. Sends an SMS + Telegram DM when a new matching listing appears.
Usage: /deal-finder (one-shot scan) — typically run from cron every 60 min.
V1 keeps the original Mac Mini search inline (state/criteria.json). V2 will move this to a state/searches.json array with one entry per search, so adding "Pelican 1535 case", "Aeron chair size B", etc. becomes a JSON edit rather than a code change.
Current Mac Mini search criteria (from state/criteria.json):
M2, M3, M4 (M1 explicitly excluded — owner asked for M2+)16 GB512 GB$50094566 (Pleasanton, CA), search radius 50 miEdit state/criteria.json to retune the Mac Mini search; v2 will lift this into a per-search config.
V1 (implemented): Craigslist (sfbay.craigslist.org/search/sss?...). Uses an honest User-Agent (Sutando-Personal-Agent/1.0) that identifies the agent rather than cosplaying a browser — Craigslist can decide whether to allow.
V2 (planned, not yet implemented):
browse.ebay.com has a local-pickup filter; HTML scraping works without auth.macos-use skill, or the browser-automation MCP) because the page is JS-rendered and gated.state/seen.json:seen.json.seen.json deterministically to the last 1000 entries (a deque(maxlen=1000) ordered by insertion — replaces the prior set-slicing trim which was non-deterministic).[Mac Mini Deal] $480 — M2 / 16GB / 512GB
Concord (35mi, local pickup)
Posted recent
https://sfbay.craigslist.org/...SMS goes via TWILIO_* env vars to OWNER_NUMBER. Telegram goes via results/proactive-deal-finder-{ts}.txt (the bridge picks it up).
bashpython3 scripts/scan.py # one-shot scan with default criteria python3 scripts/scan.py --dry-run # don't notify, print what would notify python3 scripts/scan.py --reset # clear seen.json (force re-notify everything)
Cron: every 60 min. Configured in skills/schedule-crons/crons.json as deal-finder.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 4,956 | 6,694 | +35% | 1 | 1 | 0% | 725 | 1,939 | +167% | 0 | 0 | — |
case-02 | fail→fail | 5,533 | 4,728 | -15% | 1 | 1 | 0% | 769 | 1,138 | +48% | 0 | 0 | — |
case-03 | fail→fail | 5,161 | 6,538 | +27% | 1 | 1 | 0% | 760 | 1,401 | +84% | 0 | 0 | — |
case-04 | fail→pass | 12,951 | 2,567 | -80% | 1 | 1 | 0% | 1,953 | 1,206 | -38% | 0 | 0 | — |
case-05 | fail→pass | 10,380 | 2,096 | -80% | 1 | 1 | 0% | 1,701 | 1,137 | -33% | 0 | 0 | — |
case-06 | fail→pass | 4,371 | 1,731 | -60% | 1 | 1 | 0% | 595 | 1,045 | +76% | 0 | 0 | — |
case-07 | fail→pass | 6,917 | 2,009 | -71% | 1 | 1 | 0% | 1,043 | 1,104 | +6% | 0 | 0 | — |
case-08 | fail→pass | 2,263 | 1,533 | -32% | 1 | 1 | 0% | 297 | 1,009 | +240% | 0 | 0 | — |
case-09 | fail→pass | 11,204 | 2,568 | -77% | 1 | 1 | 0% | 1,702 | 1,206 | -29% | 0 | 0 | — |
case-10 | pass→pass | 13,126 | 8,363 | -36% | 1 | 1 | 0% | 2,067 | 2,024 | -2% | 0 | 0 | — |
case-11 | fail→pass | 11,916 | 4,617 | -61% | 1 | 1 | 0% | 2,098 | 1,451 | -31% | 0 | 0 | — |
case-12 | fail→pass | 11,597 | 1,956 | -83% | 1 | 1 | 0% | 1,740 | 1,066 | -39% | 0 | 0 | — |
case-13 | fail→pass | 13,787 | 3,808 | -72% | 1 | 1 | 0% | 2,183 | 1,412 | -35% | 0 | 0 | — |
case-14 | fail→pass | 5,022 | 2,037 | -59% | 1 | 1 | 0% | 727 | 1,067 | +47% | 0 | 0 | — |
case-15 | fail→pass | 11,207 | 2,686 | -76% | 1 | 1 | 0% | 1,626 | 1,250 | -23% | 0 | 0 | — |
case-16 | fail→pass | 6,041 | 1,816 | -70% | 1 | 1 | 0% | 911 | 992 | +9% | 0 | 0 | — |
case-17 | fail→pass | 8,528 | 3,030 | -64% | 1 | 1 | 0% | 1,444 | 1,267 | -12% | 0 | 0 | — |
case-18 | fail→pass | 14,828 | 2,363 | -84% | 1 | 1 | 0% | 2,315 | 1,097 | -53% | 0 | 0 | — |
case-19 | fail→pass | 8,235 | 2,601 | -68% | 1 | 1 | 0% | 1,275 | 1,210 | -5% | 0 | 0 | — |
case-20 | pass→pass | 3,090 | 3,332 | +8% | 1 | 1 | 0% | 425 | 1,248 | +194% | 0 | 0 | — |
case-21 | pass→pass | 2,437 | 2,833 | +16% | 1 | 1 | 0% | 403 | 1,195 | +197% | 0 | 0 | — |
case-22 | pass→pass | 2,098 | 2,337 | +11% | 1 | 1 | 0% | 344 | 1,128 | +228% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +73 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.