Describe your idea
Add a configurable interaction_style setting to the core module that controls how agents ask users questions. When set to "structured" (the default), agents present numbered option lists and wait for
user selection. When set to "open", agents ask questions naturally in conversation. This is implemented via the shared activation-rules.txt template so it applies universally to all agents across all
clients.
Why is this needed?
Currently, no agents enforce structured question-asking behavior. Agents ask questions in whatever format the underlying LLM defaults to — often open-ended inline questions that can confuse users or lead
to ambiguous responses. This is problematic across the many supported clients and IDEs. A universal, configurable directive ensures consistent UX regardless of which client is used, while giving users the
choice to opt out if they prefer natural conversation. Without this, the only workaround is manually editing every agent .md file or maintaining per-client rule files — neither of which scales.
How should it work?
Agent should show claude code native AskQuestions tooling type interface whenever possible
PR
Working on this #1788
Additional context
This was identified while implementing structured question support for a BMAD 6.0.3 project across multiple supported clients. The alternative approaches considered (injecting <instructions> blocks into
each agent .md file, creating per-client rule files) all required touching 10+ files and didn't scale to new agents or clients. The activation template approach is the cleanest single-source fix.
Describe your idea
Add a configurable
interaction_stylesetting to the core module that controls how agents ask users questions. When set to "structured" (the default), agents present numbered option lists and wait foruser selection. When set to "open", agents ask questions naturally in conversation. This is implemented via the shared
activation-rules.txttemplate so it applies universally to all agents across allclients.
Why is this needed?
Currently, no agents enforce structured question-asking behavior. Agents ask questions in whatever format the underlying LLM defaults to — often open-ended inline questions that can confuse users or lead
to ambiguous responses. This is problematic across the many supported clients and IDEs. A universal, configurable directive ensures consistent UX regardless of which client is used, while giving users the
choice to opt out if they prefer natural conversation. Without this, the only workaround is manually editing every agent
.mdfile or maintaining per-client rule files — neither of which scales.How should it work?
Agent should show claude code native AskQuestions tooling type interface whenever possible
PR
Working on this #1788
Additional context
This was identified while implementing structured question support for a BMAD 6.0.3 project across multiple supported clients. The alternative approaches considered (injecting
<instructions>blocks intoeach agent
.mdfile, creating per-client rule files) all required touching 10+ files and didn't scale to new agents or clients. The activation template approach is the cleanest single-source fix.