Feature Request: Support AGENTS.md

The thread spent far less time on the mechanics of reading one Markdown filename than on what the refusal symbolizes. Several commenters see AGENTS.md support as a basic interoperability feature: teams increasingly share agent instructions across Codex, Claude Code, and open-source harnesses, and a tool that insists on its own filename forces duplicate configuration, special skills, and extra maintenance. For them, this is especially irritating because Anthropic participates in defining other AI-tool interfaces while appearing reluctant to adopt a community convention it did not originate. Others argued that the practical burden is small. A symlink or tiny CLAUDE.md shim can point Claude Code at AGENTS.md, and agents can be explicitly told to read the file. Some also cautioned against reading corporate-lock-in intent into what may simply be a hardcoded historical default. Still, workarounds do not satisfy everyone: they may miss Claude Code features such as nested-file loading, and users report that Claude sometimes disregards even the instructions it does load. Commenters broadly agreed that well-written repository guidance helps subagents move quickly and avoid project-specific traps, while bloated or poor guidance can actively degrade results. A side current mocked the scale of the dispute, arguing that AI-enabled developers are expending remarkable energy on a trivial filename convention.