Automate AI Agent Skill Generation with Bedrock Registry and AgentCore Harness

KiroHub: Generate a Kiro Skill in 60 Seconds Built With Bedrock Registry and AgentCore Harness

Alvaro Llamojha developed KiroHub to automate the creation of Agent Skills using a dataset of over 10,000 community resources. The system leverages Bedrock Registry for hybrid search and AgentCore Harness for real-world sandbox testing.

Why This Matters

Static markdown skills often fail in practice because models lack specific context or the instructions do not activate correctly in an agent’s runtime. By integrating retrieval-augmented generation through the Bedrock Registry and validating behaviors in an isolated Harness environment, developers can move from draft to production-ready skills without manual trial-and-error or compromising security environments.

Key Insights

• Hybrid search via Amazon Bedrock Registry enables semantic and keyword matching across 10,000+ Kiro resources (Llamojha, 2026).
• The search_skills tool concept allows models to synthesize specific patterns, such as PostgreSQL migration, rather than relying on general knowledge.
• Amazon Bedrock AgentCore Harness provides managed, config-based agent environments for isolated sandbox testing.
• Session-based testing in Harness allows for stateful follow-up probes to ensure skills activate correctly via frontmatter discovery.
• Automated validation uses EventBridge and Bedrock to score skills on completeness and clarity before they are published to the hub.

Working Examples

JSON-RPC call used by the generate-skill Lambda to search the Bedrock Registry via an MCP endpoint.

{"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": {"name": "search_registry_records", "arguments": {"searchQuery": "AWS Lambda error handling", "maxResults": 5}}}

Invocation payload for the AgentCore Harness to test a generated skill in an isolated environment.

{"skills": [{"path": "/workspace/skills/test-skill"}], "messages": [{"role": "user", "content": [{"text": "I need help setting up error handling for my Node.js Lambda"}]}]}

Practical Applications

• Use case: Developers can generate SKILL.md files for AWS Lambda best practices and immediately test them in a sandboxed agent environment. Pitfall: Vague frontmatter descriptions can lead to failure in skill discovery and activation.
• Use case: KiroHub uses an EventBridge-triggered Lambda to auto-validate new skills against documentation quality and completeness metrics. Pitfall: Relying solely on LLM-generated markdown without runtime validation can result in prompt injection or unexpected side-effects.

References:

• https://kirohub.dev/generate
• https://kirohub.dev
• https://docs.aws.amazon.com/bedrock-agentcore
• https://kiro.dev/docs/skills