Roo Code Feature Requests

[mrubens]

Please let us know what you'd like to see next in Roo Code!

[jtgsystems]

We would like to introduce a feature that allows us to add multiple tasks and process them sequentially until the list is complete. This could be a game-changer!

[sanderkooger]

I second that.

Having architect mode really think and set out tasks for other modes to run and later on maybe test, woudl be awesome for larger changes or adding features to your project

[sealad886]

This would actually be brilliant. A true FIFO-type queue that can be used with a pool of workers, allowing system resources to be spun up and down dynamically as well. For small projects, probably less useful, but would be great for things like:

• Creating or updating project-wide documentation
• Most phases of testing, from design to integration testing
• Init-ing projects / tools / etc (e.g. the Roo Code Memory Bank tool always starts by creating the same basic directory structure and starting files).

Just a few ideas of usefulness to start with.

I imagine a few caveats (from the obvious to the edges of my mind right now):

• Ensuring multiple tasks are able to get locks on files / codechunks they require reliably
• Ensuring sequential tasks aren't undo-ing each other: classic race condition stuff.

Possible workarounds:

• Implement grouped tasks, such that one worker picks up a group of tasks that MUST be completed in the intended order.

[misiek75]

better support for copilot models

[shellstep]

Can we get access to more models through OpenRouter? I've been wanting to use DeepSeek but cant get an official API key as the platform site has been down for days

[shellstep]

looks like this is addressed! seeing loads of options with openrouter now 🥳

[wangfh5]

I'd like to check that if the prompt caching of DeepSeek API is supported. It is quite confusing that Roo Code does not support but Cline does.

Roo Code:

Cline:

[stefanskiasan]

Hello folks Question. I would like to make the browser use with my Chrome browser. Since I have under control that e.g. I am already logged in. How can I do that?

[kimodo]

add options
when server returns an error response
Automatically retry with other API or other API Provider

[cexer]

Display tasks only related to current project: #882
Overcome model content length limitations: #883

[gitgo12]

@mrubens Generally speaking, if you want to make this app amazing, know who are your audience. Most people who actually use this application are low-code or even non coders. This means, they will suck a lot building backends. I believe if you support integration with supabase, firebase or any of these backend as services, it will make development a whole lot easier. If you notice, the most popular apps that have big audience are the ones that have backend integrations like bolt and others. The ability to give AI to create,edit,delete tables in the backend with these services would be very good. Senior developers can utilize Roo in many ways without struggle, the ones who struggle are the inexperienced programmers and non programmers who just want to build simple solutions for their use cases. Create a survey, know the audience experience level, collect answers and analyze to know what should be done next. Thanks!

[chinenvinicius]

Thats just an assumption that they are low code.

[dkamins]

No comment on the proposed integrations, but strongly disagree on assumption that Roo users are mostly low-code or non coders.

[Martech-WorkShop]

The problem with this assumption is, it's a fork in the road, do you want a tool for coders or amateurs? You can have both for a limited time, but after a while your tool becomes opinionated one way or the other.
There are dozens of tools for rookies, just a handful of tools for pros, I would much more prefer if Roo keeps going in the pro direction.
In the end, you decide if the users are low code or not with every feature and PR...
Roo is doing amazingly in the current direction, whatever that direction is, please do not change!

[HelloaZelda]

Can you add ollama apikey post feature，just like cherry studio

In this place，thank you！

[dkamins]

Some basic security and safety features regarding what is auto-approved will go a long way toward allowing users to trust the agent to do its magic. This is described and discussed here:

• Proposal (re security): Require manual approval to access files outside of project directory and other sensitive files #1063

[jazir555]

I need a way to force Roo to auto-approve all commands, even ones it hasn't seen before. It's trying to install an MCP server and debug it, but it keep making me approve the command it's attempting to troubleshoot. It's supposed to be automated but I have to babysit it ;_;

Edit: Another thing to add would be automatic OS detection. Gemini keeps trying to use && in commands and it keeps erroring when it randomly decides to try that every few commands.

[mrubens]

You can approve all by adding * to the approved command prefixes.

[Matrixmax]

Auto-complete by Tab (line auto-complete, such as Cursor Tab and continue)

[jazir555]

An IntelliJ plugin! I would love to have this integrated into PHPStorm.

[jazir555]

I'd like the ability to have Roo automatically open up a new task once the current context window limit has been reached with the same prompt and have the task keep running. I want to use Roo for debugging, and eventually the context window gets saturated and I have to start a new task to prompt it to debug again and do that multiple times until it's finished. It would be great if there was a checkbox for a repeater once the context has been saturated. A stop condition could be specified, whether that be number of loop iterations, or when a specific condition is met in the model's output (e.g. model returns no bugs found).

[jazir555]

I'd like the ability to specify rotation of different LLMs. I like to debug "by committee" and use multiple AI models since they all have different specialties and training data, so they all catch different things.

Overview

I’d love to see an automated adversarial bug testing system implemented—essentially a "code by consensus" approach, where multiple LLMs collaborate to generate, review, and refine code in an iterative loop.

Proposed Process

Code Generation – The process starts with an initial LLM generating code based on a given prompt (e.g., "Build me a recipe maker and generate a recipe for a sandwich.").

Adversarial Bug Checking – The generated code is then automatically routed to multiple LLMs (e.g., Claude and Gemini) for bug detection.

Bug Aggregation & Fixes – The detected bugs are collected, and another LLM (or the same one) implements the necessary fixes.

Iterative Improvement – This loop continues until all participating LLMs agree that the code is bug-free.

Final Debugging & Logging – The refined code is run through a debugger, with errors logged and additional debugging handled through an automated process.

This approach essentially enables bug fixing by committee, leveraging the strengths of different models to improve reliability and robustness.

Configurable Parameters

To enhance flexibility, users should be able to configure:

• Choice of LLMs – Select which model generates the initial code and which ones review for bugs.
• Bug-Fixing Implementation – Define which model applies fixes, with options for fixed selection, random rotation, or round-based rotation.
• Iteration Limit – Set a max number of bug-checking rounds (e.g., a loop of 3, 5, etc.).
• Feature Expansion Rules – Introduce additional features at predefined steps (e.g., at round 5, request and implement a new feature before continuing the bug-fixing process).
• Time & Resource Constraints – Limit the process by time (e.g., run for X minutes/hours) or API usage (e.g., cap at Y code generations).

Enhanced Workflow with Feature Expansion

To take this further, the system could incorporate stepwise feature addition:

• Round 1: Initial code generation
• Rounds 2-4: Bug checking and fixes
• Round 5: Request and implement additional feature ideas
• Rounds 6-8: Bug checking and refining the new feature
• Repeat until conditions are met

This structure enables automated iterative development, where new functionality is incrementally added and tested without manual intervention.

This approach would provide an efficient, scalable method for automated bug fixing and feature expansion, leveraging multiple LLMs in a structured workflow. It would also give users precise control over the debugging and development process while optimizing API usage.

[nnWhisperer]

It would be handy to be able to precise the permissions in a more granular way: permit write access to specific files but block others, permit accessing some websites etc...
Similar to Add Option for Directory-Level File Access Permission#365

[Blanen]

Huggingface inference providers support!

[jazir555]

Adding indication of which mode was used in a task in the history tab would be great.

[wanghsinche]

vega-lite chart display supporting

[jazir555]

When I have a task delegated by orchestrator to code mode using the Gemini 2.5 Flash model and Pro model, I get an internal server error which does not trigger an automatic retry, and I am forced to click resume task. Please add an auto retry when internal server errors or any other error occurs.

[nnWhisperer]

There is a retry permission, have you tried it?

[jazir555]

Retry is active, it retries on every error but these

[jazir555]

Please add a tab in settings to disallow certain tools (write to file, insert, diff, etc) in tasks completely.

[jazir555]

Could you please add Qwen's official API as a provider?

[BradKML]

For larger models that don't fit in the GPU?

[jazir555]

I'm not sure what you mean, Qwen's official API is cheaper.

[BradKML]

Some of Qwen is small enough to fit on the local computer's GPU, so it costs nothing.

[jazir555]

Oh then yeah, I mean for their ones that can't be run in LM Studio or Ollama on consumer GPUs.

[BradKML]

Ask in the LiteLLM GitHub repo, if that is done, all other apps would support it

[emscape]

I would love to see Figma integration. There's a plugin called Superflex that's claiming they can do it but it's....not quite there.

[tmchow]

I would like to see the equivalent of Claude Code's "Slash Commands":
https://www.anthropic.com/engineering/claude-code-best-practices#:~:text=c.%20Use%20custom%20slash%20commands

Their example of a command file is this in Claude Code:

Please analyze and fix the GitHub issue: $ARGUMENTS.

Follow these steps:

1. Use `gh issue view` to get the issue details
2. Understand the problem described in the issue
3. Search the codebase for relevant files
4. Implement the necessary changes to fix the issue
5. Write and run tests to verify the fix
6. Ensure code passes linting and type checking
7. Create a descriptive commit message
8. Push and create a PR

Remember to use the GitHub CLI (`gh`) for all GitHub-related tasks.

This is invoked in Claude Code with /project:fix-github-issues 123 for example for it to work on Github issue 123.

In RooCode this would be super powerful as you could define these commands to be quickly invoked and they could even contain instructions on what Modes/Roles to use. This would be another layer of power that would work in conjunction with everything else and allow for powerful scenarios that can help you easily interact with various MCPs, roles, etc. I think of this essentially like Macros.

I'd advocate for a folder like .roocode/commands/ just like Claude Code, then individual markdown files in that folder.

[nnWhisperer]

So, just to clarify, is what you are requesting as a feature is to create templated prompts for some tasks and call them like functions with the syntax of /functionName argument1 i.e. the slash-command syntax?
I believe that you can quickly simulate a similar behaviour, by creating a new user mode and post there the template you prepared above and change the $ARGUMENTS into ${userInput} that will dynamically change the prompt as you want. The only missing bit is, instead of the syntax, you need to click 'select mode for interaction' and select the new mode when you want to use it. I think you can just do the change in 1 minute. Let me know if you insist about the slash-command syntax and tell us the reasons.

[tmchow]

I think the power of the slash command is that you have the command running self contained essentially running in the context of the mode you're in with ability for it to be told to change modes. Much more powerful than having it be a mode itself.

Other reasons:

• Conceptually a command is different than a role/mode
• Commands can now be shared with Claude Code, even if RooCode supported other syntax within commands. Makes it easy to compare Claude Code to a RooCode app interchangeably use them with same commands.

I really think it should be a separate construct alongside user modes

[jazir555]

Can you not do that with a .roorules file?

[tmchow]

Can you not do that with a .roorules file?

It's not the same thing afaik. How would this work exactly to accomplish what I requested?

[sealad886]

Please pass reasonable log information (especially errors) back to the Roo-Code VS Code Output channel. Useful for all types of info and errors, but for the typical user probably especially important to include:

• MCP server errors, including connection errors: error title, code, AND full error text / traceback
• Provider configuration / connectivity / out of funds errors
• Code Index configuration / connectivity / out of funds errors

[sealad886]

On a related, but completely separate, note:
I'm trying to troubleshoot issues setting up an OpenAPI compatible endpoint with Code Index and running into issues. One of things I'm not sure of is whether or not OSX sandboxing features are preventing the requests from leaving VS Code at all in the first place.

I can't troubleshoot this very well because the HTTP requests don't run through VS Code's Developer Tools Network tool. Or perhaps I don't understand it well enough for it to be useful.

Tools like Wireshark are giving me too much junk for me to sort through at the moment.

Is possible to use the Developer Tools toolkit in its entirety (I know some things are logged to the Console, for example) for Roo Code?

[nnWhisperer]

What do you mean with "Developer Tools toolkit", can you give a link? I couldn't find such an MCP

[sealad886]

Sorry for the confusion there. It's not an MCP. It's built-in to VS Code (I think most chromium-based browsers in fact?). Access by opening Command Pallet and searching "toggle developer tools":

[sealad886]

Two feature requests related to the newly out-of-experimental feature Code Index:

• Reranking: Reranking Feature Request
• Throttling: Throttle the index

[sealad886]

And a Configurable TTS engine!

[Alek2077]

1. Add setting to customize how the Enter key behaves in prompt input field
2. Add ability to reset/delete parts of conversation on any AI reply, not just my own replies

[petrovichest]

I would like to optionally indicate the IPV4 proxy with AUTH through which the expansion will turn to the providers.