CI/CD Testing failures

[jasonvanwyk]

Hi All, does anyone else here also spend hours and hours just fixing and trouble shooting the failing GitHub Actions CI/CD failed tests? Every AI project i have seems wrought with conflicting dependencies, versions, requirements and especially conflicting Flake8 and Black linting setups. I have to keep correcting the models/agents when designing and building these pipelines. Any ideas? any advice? thanks in advance.

[Fcpacheco]

I’ve encountered the same issue, and there isn’t an automated solution yet. Large language models (LLMs) often fail to follow linting rules when developing code, even when provided with specific instructions. My approach is to use a Makefile to run locally, and for issues I can't address directly (e.g., running black .), I pass those to the agent to fix. I apply this method to every small part of the code, always reviewing the codebase, as incremental development can modify older files. This process typically prevents issues in the CI/CD workflow on GitHub.

P.S.: I’m using Ruff instead of Flake8, as I encountered some problems with Flake8 that the LLM couldn’t resolve.

[jasonvanwyk]

Thanks @Fcpacheco - good to know i/m not losing my mind - i havent tried Ruff before - maybe now is the the time to try it.

[bmadcode]

Im not familiar with Ruff and Flake8 - so I might be offbase what you are asking about :) At work we have been using code rabbit and its good at finding nitpicky things also in the ide - I will after a sessions of coding run code rabbit in vscode and it will offer suggestions I paste back into cursor or claude code and its good then at fixing them up. its yet another paid tool though so I have not bought it for personal use, but have been considering it. It also runs nicely in PRs and leaves comments and I assume soon will also make pr suggestions to make the fixes in the cloud, along with copilot and the other emerging cloud agents that are coming.

…

On Fri, Jun 6, 2025 at 6:19 AM Jason ***@***.***> wrote: Thanks @Fcpacheco <https://github.com/Fcpacheco> - good to know i/m not losing my mind - i havent tried Ruff before - maybe now is the the time to try it. — Reply to this email directly, view it on GitHub <#174 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BOTVGQNS6ZF2QTDW6WTY5QD3CF2KLAVCNFSM6AAAAAB6XMEKKKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGMZYHEZTAMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.*** com>

[bmadcode]

I am also trying to figure out the best way as we speak to do a potential video on it. I am loving claud code so far in combination with cursor.

Also I want to create a MCP server to add to this repo when I get some time - if you would like to contribute a pr would love to get an official one into the repo. If no one does, ill try to get to it also this weekend, but really like community contributions also! Having the MCP here will help ensure it stays in parity with changes as big improvements and some more restructures for example are coming.

[jasonvanwyk]

I'm away this weekend, but I'll try to have a MCP ready for pr by midweek next week

[fabb]

How would a BMAD MCP server work? Would it just provide tools that return instructions for different tasks, or would it execute the actual tasks itself? In the latter case, the MCP itself would need AI capability and access/keys to an LLM. In the first case it would more work like a lookup of prompts, similar to sequential-thinking or https://github.com/mettamatt/code-reasoning.

Maybe a dedicated discussion thread for MCP would be useful, so i created: #179

[bmadcode]

I suggest waiting on V4 to release this weekend - I think a lot of what ppl are looking for from an MCP server may no longer be needed after v4 is finished. I still think there is a potential use for mcp for some cases and will want to then align on if its needed and what it should be capable of.

Here is a channel for discussion of post v4 release ideas including mcp and agent frameworks https://discord.com/channels/1377115244018532404/1381071088263364638

[jasonvanwyk]

Thanks Brian - I'll think thats a good idea to wait for V4, before rushing off about MCPs etc - we can alsways iterate again -cant wait to see what you've got.