Subagents defaults are not inherited from current model

[marcociarrocca]

Love the project, it's a really cool concept IMHO!
I just installed it today (v0.2.17) and I noticed that when the deepresearch agent spawned researcher subagents, those had sonnet as the default model. That was a problem, because I am using local inference, so those subagent runs failed. I had to go in /agents and edit overrides for each subagent, setting the only available model.

[advaitpaliwal]

Fixed on main. The root problem was that bundled subagents could still resolve Pi's default user agent directory instead of Feynman's agent dir, which let pi-subagents fall back to its own pinned built-in agents. Feynman now patches pi-subagents to respect across the relevant discovery paths, and I verified the installed runtime now discovers Feynman's bundled /// from as with no pinned model. Relevant commits: dd3c076 and 01c2808.

[advaitpaliwal]

Fixed on main.

The root problem was that bundled subagents could still resolve Pi's default user agent directory instead of Feynman's agent dir, which let pi-subagents fall back to its own pinned built-in agents. Feynman now patches pi-subagents to respect PI_CODING_AGENT_DIR across the relevant discovery paths, and I verified the installed runtime now discovers Feynman's bundled researcher / reviewer / writer / verifier from .feynman/agents as source: user with no pinned model.

Relevant commits: dd3c076 and 01c2808.