v15.12.0 "Catalog Discovery and Safer Context Skills"

[15.12.0] - 2026-08-09 - "Catalog Discovery and Safer Context Skills"

Upgraded the public catalog into a practical discovery dashboard, added two focused research and planning skills, and repaired unsafe or malformed context-management guidance.

This release gives Claude Code, Cursor, Codex CLI, Gemini CLI, Antigravity, and related AI coding assistants a more usable 2,007-skill catalog. Users can now share precise catalog searches, keep a browser-local shortlist, inspect related skills, and start from clearer catalog-health signals while the underlying skill library gains bounded people research, source-aware travel planning, and stronger warnings around destructive context-agent workflows.

Start here:

• Direct skill distribution: npx agentic-awesome-skills
• Choose your tool
• Best skills by tool
• Bundles
• Workflows

Added

• Added people-data for authorized LinkedIn professional-profile and public business-contact research plus YouTube channel business-email discovery through the Agent Body MCP server, with an exact bundled tool contract and read-only boundaries.
• Added travel-planner for budget-first itinerary planning with destination constraints, dated factual sources, explicit alternatives for out-of-scope routes, and a rule self-check table in every plan.
• Added shareable fuzzy search and filters, keyboard-focused discovery, a browser-local shortlist with cross-tab synchronization and exact-ID export, catalog-health signals, and related-skill discovery to the hosted catalog.

Changed

• Updated unified-ai-gateway to distinguish the current public v0.4.3 release from the immutable v0.4.0 security-review baseline without changing its reviewed activation or inspection boundaries.
• Updated react-virtuoso to 4.18.11 in the web app and express-rate-limit to 8.6.0 in the bundled Loki example.
• Regenerated the canonical catalog, offline AAS Core data, marketplaces, bundles, Agent Plugins exports, and Codex/Claude plugin distributions for 2,007 skills.

Security

• Corrected context-agent and context-guardian to risk: critical, translated remaining guidance, and made their destructive local-write behavior and prerequisites explicit.
• Repaired malformed executable TypeScript examples in conversation-memory and context-window-management, and removed a dangling playbook reference from context-driven-development.
• Made catalog shortlist reads fail safely when browser storage is unavailable or throws, while keeping shortlist data local to the browser.

Who should care

• Users who need to search, compare, shortlist, and share exact catalog views instead of browsing more than two thousand skills manually.
• Researchers using public professional or business-contact data who need explicit authorization, read-only scope, and an exact MCP tool contract.
• Travelers who want budget-confirmed, source-aware itineraries with visible constraint checks.
• Maintainers and agents relying on context-management skills where destructive filesystem behavior must be unmistakable.

Validation

• Passed all 109 repository test groups, 173 web-app tests, web-app lint and production build, repository validation, reference validation, documentation security checks, warning-budget enforcement, and exact-head PR evidence gates.
• Confirmed root and web-app dependency audits report zero vulnerabilities, protected main CI and CodeQL are green, and canonical regeneration is idempotent.
• Verified web-app coverage at 85.04% statements, 72.95% branches, 86.37% functions, and 89.17% lines.

Credits

• @Jess-yaozu and agentbody/skills for people-data in PR #1103.
• @saudademjj and saudademjj/luopan for travel-planner in PR #1107.
• @happy520ai for the unified-ai-gateway release-note refresh in PR #1115.
• @zinzied for the original catalog-dashboard contribution in PR #1111, repaired and integrated through PR #1118.