support claude -p

[shreyjoshi2004]

All anthropic requests are currently routed through the Anthropic Console, adding support for the claude -p command would enable using the same models but with the much cheaper MAX/Pro plan. Adding an option inside config.toml for switching between claude -p and the anthropic API would be a major cost saver.

[jrswab]

So using Axe as the setup and then having it run the Claude command so you can use your max/pro plan? Am I reading that correctly?

[shreyjoshi2004]

Yes, pretty much. Would be nice to be able to integrate claude using --bare for speed, but unfortunately looks like it only allows using anthropic api key, which would self-defeating.

Another idea is project-specific agents files. You have your general purpose TOMLs (i.e. commit-msg, code-review, etc.) living at $XDG_CONFIG_HOME and then project specific agents defined within your workspace folder.

[shreyjoshi2004]

And couple other ideas for quality of life improvements:

• autogenerate a default SKILL.md
• default model version (anthropic/claude-opus --> anthropic/claude-opus-4-1-20250805)
• 'user-invocable' field for agents, in the case that you want certain specialized agents to only be agent-invocable

[jrswab]

I think claude pro/max etc is allowed with the SDK; it may be a better fit for axe to use that route. I say this as I just came across this repository yesterday using the SDK to get pro plans into OpenCode: https://github.com/rynfar/meridian