Library › Shopping & hiring
Research a purchase
Research a product purchase against the buyer's must-haves and budget, using current web sources. Returns a cited comparison table, a clear pick, and the trade-offs. Use when the user asks "what should I buy", "compare X vs Y", or "find the best X under $N".
Self-tested · live web
What "self-tested" means: our own AI agent ran this skill against the fictional demo files and we published its real output, failures included. The same agent also wrote the skill and the demo data, so this is not an independent review. Check results yourself before you rely on them.
Drop the folder into your agent's skills directory (e.g. ~/.agents/skills/ or ~/.claude/skills/), or paste SKILL.md into any chat assistant. Demo files included:
demo/brief.md
Test record
| Field | Value |
|---|---|
| Status | Self-tested (live web, Oct 13, 2026), with limits noted below |
| Version / date | v1.0 test: 2026-10-13 (no failures in this skill were scheduled for the Oct 14 fix round; re-labeled only) |
| Model / agent | Grok Bot (the same agent that wrote the skill and the demo data, so this is not an independent test) |
| Input | demo/brief.md: fictional buyer wants a tea kettle, ≤ $100, variable temp, ≥ 1.5 L |
| Method | Grok Bot followed SKILL.md using two live web searches. Sources are real; prices are as shown in search results on the date above. |
Real output
Brief checklist. Must-haves: variable temp incl. ~175°F · ≥ 1.5 L · cordless · ≤ $100. Nice-to-haves: keep-warm, minimal plastic in the water path, warranty. Deal-breaker: small gooseneck.
| Model | Price seen | Must-haves? | Strength | Weakness |
|---|---|---|---|---|
| Cuisinart CPK-17P1 PerfecTemp (1.7 L) | $100, Amazon via Wirecutter, Oct 13 | ✅ (at the budget ceiling) | Wirecutter's top pick since 2013; 6 presets incl. 175°F green tea (Allrecipes); 30-min keep-warm; 3-yr warranty | Presets only (no single-degree control); materials in the water path not verified |
| OXO Brew Adjustable Temperature Kettle (1.75 L) | Not verified | ⚠️ capacity, temp, and cordless OK; price unknown | Single-degree control 104–212°F; 30-min hold; 2-yr warranty (Bon Appétit) | No presets; one summary suggested it may exceed $100, unconfirmed |
| Chefman 1.8 L with Temp Control | Not verified | ✅ on specs (7 presets 160–212°F, keep-warm) (CoffeeGetUp) | Removable tea infuser | Only one source; no hands-on test data found |
| Cosori Gooseneck (0.8 L) | $69.99 (AislePrompt) | ❌ dropped: 0.8 L and gooseneck (deal-breaker) | — | — |
| Fellow Corvo EKG (0.9 L) | $199.95 (Fellow) | ❌ dropped: over budget, under 1.5 L | — | — |
Recommendation: Cuisinart CPK-17P1 PerfecTemp. It meets every must-have, it's the most consistently recommended across independent testers, and it has a 175°F green-tea preset. If you want exact temperatures: OXO Brew Adjustable, but confirm the price first; it may be above $100. Budget alternative: none verified at ≥ 1.5 L with variable temp. Chefman is a candidate but rests on one source.
Could not verify: live prices at checkout; whether plastic touches the water in the Cuisinart or OXO; the Chefman's real-world accuracy.
Known failures / limits
- Search snippets, not full pages. The Wirecutter page was read only as a search excerpt, not fetched in full, so details may be missing.
- No live price for OXO or Chefman. The skill correctly marked them unverified instead of guessing, but that leaves the user with only one fully priced option.
- "Minimal plastic" nice-to-have was not resolved for any shortlisted model.
- Retail links in some sources may be affiliate links; that doesn't change the pick, but users should know.
- Self-test bias: the same agent wrote the brief and ran the test.
The skill file
---
name: research-a-purchase
description: Research a product purchase against the buyer's must-haves and budget, using current web sources. Returns a cited comparison table, a clear pick, and the trade-offs. Use when the user asks "what should I buy", "compare X vs Y", or "find the best X under $N".
license: MIT
metadata:
library: TriedSkills
job: Research a purchase
version: "1.0"
---
# Research a purchase
## Inputs
A buyer brief: what it is for, budget ceiling, must-haves, nice-to-haves, deal-breakers. If any of these is missing, ask once, then proceed with stated assumptions.
## Steps
1. Restate the brief as a checklist of **must-haves** (pass/fail) and **nice-to-haves** (scored).
2. Search at least **two independent sources**: hands-on review sites that say they tested the products, plus the manufacturer page for specs. Prefer pages updated in the last 12 months.
3. Build a shortlist of 3–5 candidates that appear in more than one source.
4. Check every candidate against each must-have. **Drop** any that fail and say why. Never quietly relax a must-have.
5. For each remaining candidate, record price, key specs, and one strength and one weakness, **each with a source URL**. If sources disagree, show both values.
6. Make one recommendation, plus a "budget alternative" and an "if you can stretch" option where they exist.
7. List what you could **not** verify (e.g. live price, materials, warranty).
## Output format
Brief checklist → comparison table (Model | Price seen (source, date) | Meets must-haves? | Strength | Weakness) → recommendation → unverified items → sources.
## Guardrails
- Prices change daily. Label each one "seen on <site>, <date>" and tell the user to confirm at checkout.
- Disclose any affiliate links. Never let them change the pick.
- Don't make claims about safety or health beyond what a source says.