-
Notifications
You must be signed in to change notification settings - Fork 1.3k
Description
When following the documented OpenSpec workflow from an AI assistant, openspec validate <change-id> runs in interactive mode and blocks waiting for input. This is fine for humans, but painful for LLM tools: the model has to notice the hang, inspect the output, understand it is an interactive prompt, ask me whether to continue, and then re-run the command or simulate pressing Enter. This adds extra steps and burns tokens every time.
The CLI already has --no-interactive, but the workflow/rules that agents learn from (README / AGENTS-style instructions) do not use it for validate, so models keep calling the interactive form by default.
I've worked around this locally by editing the generated files to always use --no-interactive, but it would be great if the official agent workflows and examples also did this by default for openspec validate, especially in the propose / archive phases.
