Show HN: Skillhub Compose AI Agent

The recent Show HN post for Skillhub introduces a command-line tool designed to compose AI agent skills from disparate sources, with a specific emphasis on resolving conflicts using Claude. The project, hosted on GitHub, addresses a practical pain point: the fragmentation of agent skill definitions across different repositories and formats. The summary notes that skills.sh and agency-agents exist but cannot merge, and Skillhub’s core function—skillhub compose A B -o expert—aims to bridge that gap.

What Happened

Skillhub is a lightweight CLI utility that allows developers to combine skill definitions from multiple sources into a single, coherent agent configuration. The key innovation is its use of Claude-powered conflict resolution. When two skill sources define overlapping or contradictory behaviors (e.g., different prompt instructions for the same function), Skillhub does not simply overwrite or concatenate. Instead, it invokes Claude to analyze the conflict and produce a merged, logically consistent output. The tool is built for the terminal, suggesting a developer-first workflow where agents are composed programmatically rather than through a GUI.

Why It Matters

This project touches on a growing bottleneck in the AI agent ecosystem: skill interoperability. As more developers build specialized agents—for coding, research, data analysis, or automation—the skills that define these agents are scattered across GitHub repos, personal dotfiles, and package registries. Currently, merging two agent skill sets is a manual, error-prone process. Skillhub’s approach of using an LLM (Claude) as a synthesis engine rather than just a generation engine is notable. It treats Claude as a mediator that understands semantic intent, not just syntax. This could reduce the friction of assembling modular agent systems, moving the field closer to a “mix-and-match” model where skills are reusable components.

For AI practitioners, this has immediate relevance. If you maintain multiple agents—say, one for DevOps and one for content writing—Skillhub could let you merge their skills into a single “expert” agent without rewriting prompts. The use of Claude specifically (rather than a generic LLM) suggests the developer values Claude’s strength in handling nuanced instructions and maintaining coherence across long contexts.

Implications for AI Practitioners

• Reduced Duplication: Practitioners can stop maintaining separate skill files for similar agents. Skillhub enables a composable architecture where skills are authored once and merged as needed.
• Conflict Resolution as a Service: The tool externalizes a cognitive task—resolving contradictory instructions—to an LLM. This is a pattern we will likely see more of: using LLMs not just to generate content, but to reconcile content.
• Dependency on Claude: The tool’s core value hinges on Claude’s performance. If Claude’s conflict resolution is unreliable or biased, the merged output may degrade. Practitioners should test the tool on their own skill sets before trusting it blindly.
• Open Source Flexibility: The GitHub repo is public, so developers can fork, audit, or extend the conflict resolution logic. This transparency is crucial for trust in AI-mediated tooling.

Key Takeaways

• Skillhub introduces a novel pattern: using Claude to semantically merge conflicting agent skill definitions from multiple sources.
• It addresses a real fragmentation problem in the agent ecosystem, where skills are isolated across repositories and formats.
• AI practitioners can use this tool to reduce manual effort in composing multi-skill agents, but should validate Claude’s conflict resolution on their own data.
• The project exemplifies a broader trend of using LLMs as reconciliation engines, not just generators—a shift with significant implications for modular AI system design.