This project is mostly based on Claude Code (hence CDD). Bit part of this repo is based on prompts. On typical software projects, code can be tested with unit tests, integration tests, it can be formatted, linted, etc... all of these steps improve the reliability of the code and are usually part of a healthy CI.
However a prompt cannot be unit tested. Is there anything that we can do to guarantee that prompts work as expected? to guarantee that if a prompt is changed, the rest of the workflow keeps working as expected? Just as an example, if I change the instructions in the next-step.md it might affect how the handoff file is generated and therefore breaks the following <project>-worktree.sh step or the implementation session.
Let's investigate what's possible here and if we find something, we can implement something that improves reliability, possibly in the scope of the CI.
This project is mostly based on Claude Code (hence CDD). Bit part of this repo is based on prompts. On typical software projects, code can be tested with unit tests, integration tests, it can be formatted, linted, etc... all of these steps improve the reliability of the code and are usually part of a healthy CI.
However a prompt cannot be unit tested. Is there anything that we can do to guarantee that prompts work as expected? to guarantee that if a prompt is changed, the rest of the workflow keeps working as expected? Just as an example, if I change the instructions in the
next-step.mdit might affect how the handoff file is generated and therefore breaks the following<project>-worktree.shstep or the implementation session.Let's investigate what's possible here and if we find something, we can implement something that improves reliability, possibly in the scope of the CI.