what is athena-graphs fr?

Athena Graphs lets an AI coding agent turn a plain language goal into a structured plan made of connected steps, called a graph, instead of just looping on a task until it feels done. Each step, or node, is handled by a specialized agent, and the graph can branch into parallel research and building work, merge results back together, loop through review cycles, fall back to alternate paths, and pause at points where a human needs to approve something before work continues. In everyday use, a person does not write any graph configuration themselves. They install the plugin and simply describe what they want in a normal conversation, for example asking the agent to implement a feature, research the existing design and threat model at the same time, review the result, and pause before final approval. The bundled skill looks at the project, works out what success looks like, designs a small graph or picks a default one, starts it running in the background, and keeps the person updated as it streams progress from each node. It only stops to ask a question when the graph reaches a point that specifically needs a human decision, and it can resume later from that same checkpoint. Athena Graphs installs as a plugin for Codex or Claude Code through their plugin marketplaces, and it also works with other tools that support the Agent Skills standard and can connect to MCP, a protocol for connecting agents to external tools. It requires Python 3.10 or newer and the uv package manager, but does not require the user to install any Python packages directly, since the plugin sets up its own environment automatically the first time it runs. Under the hood, the plugin exposes a set of MCP tools for starting a graph, checking its status, streaming its output, resuming it after a pause, fetching its final result, and rendering it as a diagram. Runs are saved to disk so they can be inspected or resumed later. Advanced users can also write a graph directly as a JSON structure describing nodes and the conditions that connect them, though the skill normally generates this automatically. The project is written in Python and released under the MIT license.