Styleguide is a living document and it will never be "100% complete". Regular improvements are being made to it manually, but it requires a lot of effort from humans to manage. Creating well defined issues for every single minor thing that is currently "bad" is very boring, and even more than that - people cannot keep the whole styleguide in their head at all times when reviewing PRs, so in many cases they would simply miss potential things to improve in styleguide.
AI could help with this problem long-term. Suggested workflow:
- after every merge of PR we run an action, giving agent read access to all issues & PRs
- in this action we ask agent to go through all the review comments in that given PR, all the "fix" commits, and human feedback on AI-review findings
- after gathering all the context it starts thinking about "how styleguide could be improved", and checking out other PRs, issues, commits - looking for similar cases
- if any specific things to improve in styleguide were found - agent opens a well described issue explaining the problem, linking examples, suggesting improvements, and then also opens a PR resolving this issue
The prompts for this thing should be strict enough to not let AI suggest dozens of improvements after every single PR.
Styleguide is a living document and it will never be "100% complete". Regular improvements are being made to it manually, but it requires a lot of effort from humans to manage. Creating well defined issues for every single minor thing that is currently "bad" is very boring, and even more than that - people cannot keep the whole styleguide in their head at all times when reviewing PRs, so in many cases they would simply miss potential things to improve in styleguide.
AI could help with this problem long-term. Suggested workflow:
The prompts for this thing should be strict enough to not let AI suggest dozens of improvements after every single PR.