22 Agent Skills for Amazon sellers and advertisers: Sponsored Products and Brands audits, search-term mining, budget pacing, DSP and Amazon Marketing Cloud, reporting, listing optimization and compliance, product images, repricing, FBA inbound, stockout, refunds, buyer messaging, and profitability. Built for Kuudo's hosted Amazon Ads and Selling Partner MCP servers; price, budget, listing, and shipment writes are previewed and wait for explicit approval.
68
93%
Does it follow best practices?
Impact
66%
1.65xAverage score across 22 eval scenarios
Medium
Suggest reviewing before use
#!/usr/bin/env python3
"""
Re-run the documented worked examples against the scripts.
Both reference docs say "re-run this case after any change." This is that command.
The fixtures in fixtures/ carry their own `expected` block, so the numbers quoted in
references/formulas.md and references/period-pnl.md are checked against the code
rather than trusted, and a change that silently moves a fee model or a checkpoint
fails here instead of in someone's margin.
python scripts/validate_fixtures.py
"""
import json
import pathlib
import sys
sys.path.insert(0, str(pathlib.Path(__file__).parent))
import margin # noqa: E402
import pnl # noqa: E402
FIXTURES = pathlib.Path(__file__).parent / "fixtures"
TOL = 5e-3 # half a cent, or half a tenth of a point
def compare(label, got, want, tol=TOL):
ok = got is not None and abs(got - want) <= tol
print(f" {'PASS' if ok else 'FAIL'} {label:<30} expected {want:>12,.4f} got "
f"{'None' if got is None else format(got, '>12,.4f')}")
return ok
def check_per_unit(path, doc):
exp = doc["expected"]
r = margin.run(doc)
ok = True
for key, want in exp.items():
if key == "loss_zones_count":
got = len(r["loss_zones"])
hit = got == want
print(f" {'PASS' if hit else 'FAIL'} {key:<30} expected {want:>12} got {got:>12}")
ok &= hit
elif key == "warning_contains":
hit = any(want in w for w in r["warnings"])
print(f" {'PASS' if hit else 'FAIL'} {key:<30} expected {want!r} in warnings")
ok &= hit
else:
ok &= compare(key, r.get(key), want)
return ok
def check_period(path, doc):
exp = doc["expected"]
r = pnl.analyse(doc)
ok = compare("net_revenue", r["checkpoints"]["net_revenue"], exp["net_revenue"], 0.05)
for key in ("product_margin_pct", "channel_margin_pct", "growth_margin_pct",
"net_operating_profit_pct"):
ok &= compare(key, r["margins"][key], exp[key], 5e-4)
ok &= compare("tacos", r["tacos"], exp["tacos"], 5e-4)
return ok
def main():
all_ok = True
for path in sorted(FIXTURES.glob("*.json")):
doc = json.load(open(path))
print(f"\n{path.name} — {doc.get('_case', '')}")
runner = check_period if path.name.startswith("period-") else check_per_unit
all_ok &= runner(path, doc)
print("\nAll fixtures pass." if all_ok else "\nFIXTURE FAILURES — do not ship.")
return 0 if all_ok else 1
if __name__ == "__main__":
sys.exit(main()).tessl-plugin
skills
amazon-ads-accounts
amazon-ads-budget-pacing
amazon-ads-create-dsp-campaign
amazon-ads-creative-assets
amazon-ads-marketing-cloud
amazon-ads-reporting
assets
templates
sb
references
tests
amazon-ads-sb-performance-review
amazon-ads-sp-campaign-auditor
amazon-ads-sp-search-term-explorer
amazon-product-image
amazon-profitability-calculator
amazon-sp-fba-inbound
amazon-sp-listing-compliance
amazon-sp-listing-optimizer
amazon-sp-messaging-api
amazon-sp-refund-return-monitor
amazon-sp-repricing
amazon-sp-sales-traffic-analyzer
amazon-sp-sqp-asin-ngram
amazon-sp-stockout-prevention
connect-to-kuudo
using-kuudo