Allow slash commands to be defined as Markdown files under `.jupyter/commands`

[dlqqq]

Claude Code and other agentic assistants allow users to define new slash commands in natural language stored in files like AGENTS.md.

There are two options for implementing this feature:

1. Make custom slash commands be persona-specific. This would mean that every persona needs to provide its own way of reading AGENTS.md, providing slash command completions, and handling slash commands routed to it like @Jupyternaut /custom-slash-command.

2. Implement this capability inside jupyter_ai_chat_commands. Perhaps, this package could detect the default persona in the current chat by calling something inside jupyter_ai_router. This would allow jupyter_ai_chat_commands to call the persona directly to generate a list of slash commands defined in AGENTS.md, and automatically provide completions for those slash commands through the UI. This would require a new get_custom_slash_commands() method on BasePersona.

[3coins]

@dlqqq
I believe AGENTS.md support is going to be different from custom slash commands. The way custom slash commands seem to work is via a set of markdown files under the .claude/commands dir which is different from the overall project level memory provided by CLAUDE.md. See here for more info on how this works. Let me know if I missed anything.

[dlqqq]

@3coins Thanks! Renamed the issue title accordingly