The aim of this project is to provide a boilerplate Skill that developers can plug into coding assistants (like OpenCode) to generate agentic workflows. It uses the Microsoft agent-framework library and provides a functional Terminal User Interface (TUI) as the baseline scaffold. When working with resource-constrained local models, auto-generated agents often lack the architecture to isolate context, impose loop limitations, and handle file reading efficiently.
This project provides a template that includes quotas, contextual boundaries, and sub-agent delegation functionality to help generated agents execute within the bounds of local hardware.
This skill is designed to help Local AI Coding Agents build Local Agent Workflows.
- Scaffolding LLM Compatibility: Tested via OpenCode utilizing the Qwen 3.5 (122B) model to comprehend the skill rules and construct applications using the
basic-tui-agentscaffold. - Agent Workflow LLM Compatibility: The generated agents and workflows have been validated running locally on the Qwen 3.5 model family.
- Hardware Profile: Tested on local consumer hardware, specifically AMD Ryzen AI HX 370 (Strix Halo) and AMD Ryzen 9 7900 systems.
If you are using a coding agent like OpenCode, place this repository folder into your active skills directory (e.g., ~/.opencode/skills/). Following the embedded SKILL.md rules, the coding agent will know how to bootstrap and template your local agent applications using Microsoft's Agent-Framework.
SKILL.md: The primary directive file loaded by the coding assistant.resources/docs/: The architectural rulebooks outlining context isolation, UI constraints, and LLM prompt engineering strategies for local execution.examples/basic-tui-agent/: The read-only baseline scaffold that coding assistants copy when beginning a new local agent project.
Read the built-in AGENTS.md for more details regarding how the specific execution loops were designed to navigate hardware limits.