Claude Code

“Hermes ships a claude-code skill that hands coding work (features, PRs) to the Claude Code CLI, and a codex one that does the same for Codex. so a setup a few people here run is Hermes as the main agent, keeping your memory, preferences and skills across sessions and chat apps, with Claude Code doing the heavy coding when Hermes sends it there.”

“Use Hermes primarily but give it access to Claude Code so it can delegate coding work there. Works great for me.”

“The bottom line is the same as OP's: if you're running agents headless on a Pro plan, budget your 5-hour limit accordingly. It's gonna hurt. Keep the big jobs interactive to save your sanity (and your subscription).”

“I am using claude code on cli / desktop app / Xcode / VSCode (sometimes) but %60 on cli.”

“In Claude Code you can see what is going on better.”

“If I use Claude code app and ask it to create a sub agent to do XYZ, is that headless?”

“Measured it: headless Claude Code (Agent SDK / claude -p) eats ~3x more of the 5-hour limit per token than interactive use (vscode)”

“Claude Code is a better harness for coding.”

“I’ve been using claude code all this while and I heard that hermes is better at learning how you work and it can automatically create necessary skills based on ur workflow u don’t have to manually do it like in claude code/codex.”

“the line at the bottom of the terminal showing context left, model, git branch and so on. it's a script you point claude code at: it hands your script a json payload on stdin every render, your script prints one line, that line is the status bar.”

“My team uses Claude Code, while I use GitHub Copilot in VS Code.”

“The best thing is that it runs with agent sdk which is basically claude code. So you can add any feature on your mind on top of that”

“The extension feels much easier for chats and reviewing changes, but the CLI seems better for giving Claude a bigger task and letting it work more independently.”

“I am trying to understand if this is something different than what I have been doing in the CLI for a while now, with my main session launching multiple teammates to accomplish tasks its been giving, and controlling them, and returning results. Is this just Desktop / Web catching up?”

“Claude code on Windows is notorious for this. The reason is that it uses git bash for the terminal shell by default. This terminal shell does not set up proper termination handling for child processes.”

“I have a skill for cross-session-messaging (it's so new most models don't know they can do this!) and handoffs (with continue, fresh, greenfield, brownfield args to hand off, as well as args for model, effort, branch + worktree setups). It's pretty sweet!”

“How to connect claude code terminal and claude.ai?”

“Claude Code can even do that for you. Just lately it informed me that it couldn't run unit tests because other Claude Code processes and Elden Ring were using too much memory.”

“To be honest, what's eating all the memory is vs code with claude code. I check the same with pi agent few days ago and it was like a huge less consumption”

“Right now, well, I hate Claude Code, I simply can't stand it's interface and how it does everything, it's not transparent and it's awkward. So I use opencode instead. If Claude Code lets me fix the interface, then I might actually use it.”

“Yea, crazy that claude code can’t solve this one problem!”

“Does in Claude Code also mean CLI?”

“Claude Code is getting native AGENTS.md support!”

“this is a claude code agent running in the claude code harness”

“After using Claude Code across multiple projects, I kept ending up with random ports/services running days later.”