Work / Developer tooling & evaluation
Experimental projectDevAgent CLI
A small AI development workflow you can inspect.
An experimental Python CLI that separates commands, prompts, file tools, orchestration, and model calls, with traces that make each workflow step easier to inspect.
A readable workflow
I built DevAgent around a compact path: CLI → Orchestrator → Tools → LLM → Output. Separating those responsibilities makes it easier to follow how a file becomes a prompt and how a model response becomes something to review.
What the repository contains
The public repository contains three commands with a shared orchestration path. Command-specific prompts give each task its own output shape, while trace entries show the steps leading to the result.
- explain: request an explanation of a target file.
- fix: request suggested changes using a defined Markdown response structure.
- gen-api: generate an API draft from a requirement file.
- Trace steps for file reads, directory summaries, prompt selection, model calls, and optional output writing.
Current status and limits
DevAgent is an experimental implementation. Its traces cover workflow steps and directory listings; they do not demonstrate comprehensive repository understanding. The file tools perform local reads and writes without a verified security sandbox.
The linked source supports the workflow described here. Model quality and runtime reliability still need evaluation before the tool can be treated as ready for production use.