Replies: 6 comments · 5 replies
-
|
In my testing I've seen some models maintain a task list and others not. (I'd have to snag a screenshot next time I see it happen) How have you handled tasks in your implementation? Have you modified the core prompt or are you adding a tool(s)? |
Beta Was this translation helpful? Give feedback.
-
|
^ Might be interesting to look at this thread - handles a similar need |
Beta Was this translation helpful? Give feedback.
-
|
Hey, for the sake of clarifying of the technical details, I've put up my implementation as a draft PR, and it is ready to try out: #746 But here's the quick scoop:
|
Beta Was this translation helpful? Give feedback.
-
|
How far do you think we are from having fully operable "Flow" tooling similar to Claude Flow? #786 |
Beta Was this translation helpful? Give feedback.
-
|
hey @BradKML I'm not actually really familiar with Claude Flow! What you think is needed to get that working? |
Beta Was this translation helpful? Give feedback.
-
|
Mostly about getting rid of dependencies in API keys (same for Claude Flow AND two iterations of Gemini Flow), other than that it should be just handling MCP calls and making sure there are no incompatibilities e.g. GPT-OSS Harmony format, OpenAI-Compatible API vs Anthropic API format wars. Ideally, if we enhance Crush with itself (like the OpenHands and Claude Flow playbooks), then it should be able to "absorb" the features of Claude Flow and other project tools into its native system. We don't just need TODO lists but also bug-tracking mechanisms AND problem tracing. LocalIssues and "vibe check MCP" are a start, but not sufficient as part of the whole vibe-check system. https://github.com/yinwm/LocalIssues https://github.com/PV-Bhat/vibe-check-mcp-server Bonus points go to vector DB and documentation tracking. Pick some SOTA multilingual code/language embedding, and it should be okay as an option. |
Beta Was this translation helpful? Give feedback.
-
|
Just a small note from the Charm team that we hear you and this is a feature we'd like to see implemented soon. |
Beta Was this translation helpful? Give feedback.
-
|
Please Kanban the todolist for this project, so the users can anticipate what could be coming next |
Beta Was this translation helpful? Give feedback.
-
I find tools like https://www.task-master.dev super helpful when collaborating with LLM. You and the LLM can update the tasks independently at your own convenience. It isn't always easy have it use it and show progress consistently your PR would be helpful for that. I don't see why a native (to Crush) + MCP solution could not coexist. Crush could implement the UI and tooling for interact with it. MCPs can be used to have more complex task management. |
Beta Was this translation helpful? Give feedback.
-
|
see also: #746 |
Beta Was this translation helpful? Give feedback.
-
|
FWIW closed #746 so we can centralize discussion here. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hey Charm team,
I'm working towards moving from Claude Code (which is both closed source and coupled with a single model vendor) to Crush. However, I find myself missing its task tracking tool feature, where the model is given access to a simple for keeping track of TODO items as it works, scoped to the current session and project PWD. Just like humans, the model benefits¹ from keeping a notebook on the side as a scratchpad to tactically plan ahead, track its work and maintain focus.
I've actually already implemented this in Crush, but the contribution guide suggests opening a discussion before opening a PR to discuss the value of the feature. Here's a screenshot of it in action:
And unlike CC, Crush's snazzy sidebar was a natural place to show a live list of the TODOs!
(Incidentally, as a follow up to this feature, I'm hoping to also implement an equivalent to CC's Planning Mode, where tools that mutate state/files are disabled and the model is prompted to produce a short markdown spec document, before making any TODOs or beginning any changes.)
¹ I'll be honest, I actually have no real way to truly validate this assumption that TODO tools aid models, but given that other Agents have had this feature for a while, it seems like a pretty easy bet to me.
Beta Was this translation helpful? Give feedback.
All reactions