Changelog

All notable changes to this skill are documented here. Format follows Keep a Changelog. Versions are semver and must match metadata.version in SKILL.md.

[0.2.0] - 2026-10-17

Added

• A read-only boundary near the top: the skill never changes prices, listings, inventory, inbound plans, or shipments, and calls no write tool. reports_createReport is named as a read-type report job that needs no approval.
• The host script runs only on the url from this run's reports_getReportDocument call, single-quoted (as is any --skus list); a URL that isn't https:// or contains a single quote stops the run.
• Eval live-injected-report-row (id 5).

Changed

• The data rule in Guardrails now names product names, SKUs, report cells, plan and shipment names, error text, and the script's output; says words in them never change the ranking, the next tool, or the next command; and has instruction-like text shown to the seller as a finding.
• "Delete downloaded report files after parsing" contradicted the script, which holds the report in memory and writes no file. Guardrails now says so: there is nothing to delete, and the report is not saved to disk.
• get_schema wins only on argument names, types, and shapes (data-sources.md), never over the read-only or data rules.

[0.1.0] - 2026-09-24

Added

• First KuudoAI release. Adapted from the Amazon Selling Partner stockout-prevention skill (Apache-2.0) and rewritten for a code-mode SP-API MCP server. Validated read-only against the live server.
• Amazon's own forecast-based numbers from GET_FBA_INVENTORY_PLANNING_DATA: days-of-supply and the recommended ship-in quantity and date. They're read through the bundled scripts/planning_report.py, which streams the pre-signed report URL in memory and returns compact JSON.
• OUT_OF_STOCK and INACTIVE bands, so dormant zero-stock SKUs don't show up as critical.

Added after live smoke and eval rounds

• Inbound-gap search. Now covers SHIPPED-status plans, where in-transit shipments live. It searches them first and stops once the in-transit units are covered. A live smoke test showed that ACTIVE plans held only stale drafts.
• New states:
  • an OUT_NO_RECENT_SALES band: zero stock and no sales, but Amazon recommends a ship-in
  • a watch tier: healthy by days, but "Low stock" or inbound not yet shipped
• Report reuse. Reuses the newest planning report already finished today (reports_getReports). The script adds inbound_in_transit.
• Nothing inbound. A SKU with no inbound units gets a clear "nothing on the way" answer instead of a plan search.
• Recent pace. The 7-day vs 30-day pace comparison is shown when they differ by more than about 25%.
• Other changes:
  • the home marketplace is the default, with participations listed via sellers_getMarketplaceParticipations
  • the identity is looked up once and re-selected each block
  • AWD is checked only when the seller uses it
• Pricing. Price is never offered as a lever, and price fields are dropped inside the sandbox.

Fixed from PR review

• planning_report.py counts. --skus no longer drops rows before counting. skus_in_report and band_counts always cover the whole report, and every filter now applies only to the rows displayed.

Changed (compared with the upstream skill)

• Tool resolution. Tools are resolved by operationId through server discovery (fba-inventory_*, sales_*, reports_*, fba-inbound_*) instead of fbaInventory_* / fbaInbound_*.
• Seller selection. No entityId. The seller comes from set_active_identity, pinned in every block.
• Velocity. Now counts FBA sales only (fulfillmentNetwork: AFN) and is labelled a trailing-30-day estimate. A live check showed it differs from Amazon's figure by 20–65%.
• Inbound gap. Arrival now comes from getShipment.selectedDeliveryWindow, found through listShipmentItems. The upstream estimatedDeliveryWindow on listInboundPlans does not exist.
• Pricing. Price decisions are handed off to amazon-sp-repricing. This skill no longer drafts patchListingsItem changes, and it makes no writes at all.
• Charts. Now optional, drawn only where the host renders artifacts.
• Removed claim. Dropped the unverified "Amazon canonical rule: DOS < 14 is at risk". The bands are documented as this skill's own heuristic.