anntnzrb amz-live
Read-only Amazon catalog search through the bundled skill-local CLI. Use it for Amazon product discovery, price comparison, shortlist generation, cheap-but-decent recommendations, connector/type-specific cable hunts, structured output, delivery-location reruns, and process/RPC integration.
Entry points
- From the skill root:
uv run --script <skill-dir>/scripts/cli.py ... - If
SKILLS_DIRis set:uv run --script "$SKILLS_DIR/amz-live/scripts/cli.py" ... - Otherwise resolve the skill directory, then run
uv run --script <skill-dir>/scripts/cli.py ... - For process integration:
uv run --script <skill-dir>/scripts/cli.py --mode rpc
Core rules
- Prefer
--llm-jsonunless the user explicitly wants human text - Start cheap; avoid
--detailson a broad first pass unless the shortlist is already tiny - Add
--zipearly when delivery locality matters - Add
--details --detail-limit 2only for finalists - Add
--scoringwhen the user wants “best”, “good enough”, “not trash”, “value”, or ranking help - Answer from envelope fields, not vibes
- Do not dump raw JSON unless asked
Required follow-up reads
| Need | Read | When |
|---|---|---|
| Maintainer contract | README.md |
Before changing CLI, parser, tests, or machine-readable behavior |
| Shortest correct command | references/cheatsheet.md |
For discovery, shortlist, details, scoring, zip-aware search, fixture parsing, or trusted fields |
| Multi-step routing | references/workflows.md |
For recommendations, accessory filtering, finalist validation, locality reruns, enrichment, or answer shape |
| Operational contract | references/operational-contract.md |
For output modes, exact controls, locality, enrichment, scoring, accessory heuristics, failures, or evidence fields |
| Process integration | references/rpc.md |
For --mode rpc, request/response schema, zipCode, or query.zip_code |
| Recovery | references/troubleshooting.md |
For blocked fetches, sparse details, locale drift, parser drift, conflicting fields, or fixture debugging |
Workflow
- Identify the task shape:
- discovery
- shortlist/ranking
- detail validation
- programmatic integration
- blocked fetch debugging
- location-sensitive reruns
- Run an
--llm-jsondiscovery pass with--limit 10; add--zipif locality matters - Tighten with
--max-price,--min-rating,--include,--exclude,--title-contains,--badge,--zip,--page,--pages, or--amazon-sort - For finalists, rerun with
--details --detail-limit 2 --scoring --limit 5 - Inspect top 3-5 results, then summarize price, rating, review count, brand, merchant trust when available, delivery-locality caveat when using
--zip, and mismatch/risk
Reference routing
Use the Required follow-up reads table near the top of this file; do not preload references for routine searches.