This suggests adding a `QWEN.md` in the repo for agents instructions.
Where are we with `AGENTS.md`? In a team repo it's getting ridiculous to have a duplicate markdown file for every agent out there.
Can't these hyper-advanced-super-duper tools discover what UNIX tools since circa 1970 knew, and just have a flag/configuration setting pointing them to the config file location? Excuse me if they already do :-)
In which case you'd have 1 markdown file and at least for the ones that are invoked via the CLI, just set up a Makefile entry point that leads them to the correct location.
It would be funny to write conflicting instructions on these, and then unleash different coding agents on the same repo in parallel, and see which one of them first identifies the interference from the others and rewrites their instructions to align with its own.
> This node.js CLI tool processes CLAUDE.md files with hierarchical collection and recursive @-import resolution. Walks directory tree from current to ~/.claude/, collecting all CLAUDE.md files and processing them with file import resolution. Saves processed context files with resolved imports next to the original CLAUDE.md files or in a specific location (configurable).
I mostly use Claude Code, but every now and then go with Gemini, and having to maintain two sets of (hierarchical) instructions was annoying. And then opencode showed up, which meant yet another tool I wanted to try out and …well.
Maybe there could be an agent that is in charge of this and it's trained to automatically create a file for any new agent, it could even temporarily delete local copies of MD files that no agents are using at the moment to free the visual clutter when navigating the repo.
