Agent Skill Harbor: GitHub-native skill management for AI agent teams
What Agent Skill Harbor does
Agent Skill Harbor addresses the gap between public skill discovery and personal skill management by providing a team-focused platform for AI agent skills. It's designed as GitHub-native, DB-less, and serverless because skills are primarily text artifacts that fit naturally in Git workflows.
Key features from the source
- Collects skills from GitHub repositories
- Tracks provenance of skills
- Supports governance and safety checks
- Publishes a static catalog site using GitHub Actions and GitHub Pages
- Open-source (OSS) platform
Technical approach and context
The creator notes that while MCP (Model Context Protocol) prompt delivery could provide dynamic skill distribution in the future, Git-native approaches are currently more practical because:
- Skills are mostly authored and reviewed in Git
- Teams need provenance and governance around skills
- Tool support for MCP prompt delivery is still incomplete
Agent Skill Harbor is positioned as addressing organizational needs like collection, cataloging, provenance, governance, and safety on top of individual skill packaging approaches.
Community discussion points
Hacker News commenters discussed several related topics:
- MCP support could enable dynamic skill feeds without sync workflows
- Standardized CLI skills protocols similar to --help for agent/human workflows
- Whether skill management should extend beyond prompts to include MCP, commands, hooks, and rules
- Debate about whether skills are just text (prompts and scripts) or can include binaries
- Discussion about static vs. dynamic skill delivery approaches
The demo is available at https://skill-mill.github.io/agent-skill-harbor-demo/ and the repository at https://github.com/skill-mill/agent-skill-harbor.
📖 Read the full source: HN AI Agents
👀 See Also
cortex-engine MCP server adds persistent memory and multi-agent support
cortex-engine v0.4.0 is an open-source MCP server that gives AI agents persistent long-term memory with tools like observe(), query(), believe(), and dream(). It now supports multiple agents with isolated memory namespaces.
Why AI Coding Agents Churn Out Slop After 20 Turns: Context Blindness
A deep audit of API logs reveals Cursor and Claude Code aren't getting dumber — they suffocate on bloated context windows filled with noise, causing architectural destruction.
Learning-Kit: A Claude Code Plugin for Codebase Onboarding and Exploration
Learning-kit is a free Claude Code plugin that analyzes repositories to generate structured learning plans and interactive tutorials. It helps developers understand unfamiliar codebases before making changes, with configurable enforcement modes and progress tracking.
Hybrid search with RRF improves AI memory system over pure vector search
An open-source AI memory system using PostgreSQL with pgvector found pure vector search insufficient for exact matches, so it added full-text search and merged results using Reciprocal Rank Fusion (RRF) with k=60, plus query enrichment via tokenizer.