[19.6.0] - 2026-10-13 - "Focused Plugins, Safer Workflows And Complete Agent Setup"
Adds 2 catalog skills and expands 21 focused plugins with 48 useful memberships. Ships 2,689 skills, with complete Codex/Claude setup guidance and reviewed fixes to API, testing, accessibility, analytics and agent-evaluation examples.
Added
- api-contract-testing — provider/consumer runtime contract checks, invalid-response fixtures and version compatibility, using the project's existing tools.
- mcd-secret-menu-lab — community recipe planning for McDonald's China through an already authorized MCP connection, with current-price/nutrition checks and explicit coupon/order boundaries. Installing the skill does not install an MCP server or bundle the upstream quiz/recipe application.
Changed
- Expanded all 21 specialized plugins to 6–13 skills while keeping plugin IDs stable. Security adds threat modeling and skill audits; QA/API add contracts and bounded retries; other packs add focused UI, document, analytics, incident handoff, pricing, onboarding and privacy workflows.
- Updated riffkit with Adapt, Swap and original-idea video workflows, quote-before-render approval, current free-plan guidance and retry-cost caveats.
- Made the active reverse-engineer-anything upstream location discoverable while retaining historical import provenance and the reviewed local references.
- Completed Codex and Claude setup guides: approved configuration with backups, connection checks, exact skill selection, immutable plan preview and installer handoff.
- Improved skill pages and navigation: readable articles, direct SKILL.md copying, retained catalog filters and clearer related-skill context.
Fixed
- Replaced 18 template or incomplete descriptions and rejected known empty/template descriptions in validation.
- Corrected asynchronous test assertions, Python negative fixtures, webhook rejection under optimization, API retry statuses and provider wait budgets, alert units, contrast colors, scoring/savings arithmetic and evaluator JSON/calibration guidance.
- Removed unavailable resource promises, unpinned accessibility-tool downloads and unsupported operational claims from the reviewed cohort.
- Made plugin membership tests independent per pack so complete coverage fits normal CI deadlines without weakening assertions.
Review scope
The complete catalog metadata and all 21 specialized compositions were examined. Deep content review covered the selected addition cohort, changed skill trees and this three-PR maintenance batch; it is not a claim that every catalog procedure ran successfully against live services. Automated semantic review and manual exact-head review remain distinct.
Thanks
Thanks to @Reak-Z, @owengu-ai and @morluto for this maintenance batch, and to the upstream authors of the existing skills selected for the expanded plugins.