HVE on Claude Code #2436
Replies: 4 comments 1 reply
Follow-upBy way of some basic instructions, Claude Code custom agents aren't selectable in a dropdown (or dropup?) menu like in GHCP chat. To select the different agents, one must type It will prompt you for a project slug to use and then begin the coaching session. I've been testing it out this past week and it's functioning (so far) as intended. Sonnet 5 is great. I'm looking forward to comparing my client work in Claude Code with my Avanade work in VS Code GHCP using the GPT 5.6 models (starting with Terra, as it's the most analogous to Sonnet). |
|
Tried it out. Works like a charm. Thanks! Would have to think of a way to sync it to future HVE-Core releases. |
|
Glad it works for you! Future-sync is definitely something to think about. |
|
I manually copied many of the HVE-core agents/skills to be Claude agents on an engagement that I am wrapping up now. I think the way Claude works overall that we probably would want to have things like the researcher, planner, implementor be claude skills rather than agents. I had lots of problems when the claude agent would try to use these as sub agents and then passing messages back and forth with me. Especially the PR-review agent, I had times where the Claude agent would ask me if I wanted to approve or reject a comment, I would tell it my answer and it would reply to the pr-review sub-agent. The sub-agent would then respond with something like "No, unless I hear directly from the user I will not proceed." The agents would get into arguments over this and waste tokens! |
|
Thanks for the testing and feedback! Valuable experience. That's not a bad idea. I'll feed that in and see what can be done. The 4.0.0 release is also VERY near, which will be more skill-heavy than agent-focused. It might be good to take a second look once the new release drops. |
Uh oh!
There was an error while loading. Please reload this page.
Good afternoon, folks!
After our awesome workshop, I took a week of PTO, and have just this week gotten back into the swing of things. Today, I got HVE core working successfully within the Claude Code framework. I'll share the basic insights I gathered along the way, and why it's not a simple plug-and-play.
The basic issue is that Claude Code does not directly read the various agents, skills, and prompts straight from the VS Code extension (or at least I didn't figure out how to make it do that and wasn't willing to spend more than an hour on it). Claude instead saves these system-wide tools in the
$LOCALUSER/.claude/directory. What I ended up doing was having Claude Code examine the repository (because as we pointed out during said workshop, who reads code anymore?) and used Prompt Fu to guide Claude toward repackaging the agents, skills, and prompts in such a way that it is more suite to be directly consumed by Claude Code.There are some slight differences as well between the UX (how Claude prompts users for answers to questions - Claude is more text-based than GHCP) and some small changes in how prompts hand off to agents/skills. I had to change a few files, but not significantly, except for 3 having to be combined prompt/agent files.
I've been using it now within Claude Code and it's working well. No major hiccups so far. I've also uploaded a .zip file that contains what would go in your
$LOCALUSER/.claude/, directory. I've also created two other files: one is a.cmdfile that should be a double-click. That'll start up the paired.ps1file, which should auto-install all the required files and append the required info to your.CLAUDE.mdfile to empower HVE. Unfortunately, GitHub won't let me upload those two files to the discussion here, so all you get is the zip file. If any of you folks have access to Claude Code, I'd welcome your feedback.Keep in mind that the contents of the zip's
.CLAUDE.mdfile should be copied and appended to your existing.CLAUDE.md, not overwrite it.This version is based on the latest pre-release of HVE Core - All.
hve-core-claude-code.zip
All reactions