Claude Code Teams, Nested Skills, and Privacy Fixes

Lachlan Reed

Brought to you by Jellypod AI, we are jumping straight into the terminal today because my mind is absolutely blown by what Anthropic is doing with the latest Claude Code CLI. I just spun up a local build using `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` and the way it handles multi-agent orchestration now is incredibly slick.

James Turner

Oh, that experimental teams flag is a massive paradigm shift. [excited] It completely kills the old manual overhead. You don't have to run verbose commands like `claude agent create` or manually manage JSON state files anymore to get agents talking to each other. Instead, you get this single implicit team where you can literally just spawn sub-agents directly by name right in your prompt.

Lachlan Reed

Exactly, it's flat out brilliant. No dramas setting up complex orchestrator files. But where this really saved my skin this week was on a massive monorepo I've been wrestling with. They've introduced nested skill resolution, so the CLI now automatically crawls your project structure and registers any `.claude/skills/` directory it finds in your subfolders.

James Turner

Wait, [genuinely surprised] if it's auto-detecting skills across a giant monorepo, how does it handle name collisions? Like, if you have five different microservices that all define a custom `deploy` skill, doesn't the global namespace just collapse?

Lachlan Reed

You'd reckon so, but they actually solved it quite elegantly. It automatically prefixes the skill with its subdirectory path. So instead of a collision, your API deployment skill gets loaded dynamically as `packages/api:deploy`.

James Turner

Ah, `packages/api:deploy`! [thoughtfully] That is incredibly clean because it mirrors the exact syntax developers already use in monorepo task runners like Turborepo or Nx. It means the LLM can resolve the correct context natively without you having to go in and manually rename twenty different skill files just to keep Claude from tripping over its own feet.

Lachlan Reed

Spot on, it just fits right into your existing mental model. And they've also added some really nice ergonomic tuning to the CLI itself. Instead of having to open up and edit global JSON configs in some hidden directory, you can now do instant, on-the-fly terminal adjustments using the new `/config key=value` syntax.

James Turner

Finally! [laughs] No more breaking your flow to look up where the config file is hidden on macOS versus Linux. You just type `/config theme=dark` or whatever right in the active session.

Lachlan Reed

Too right, mate. And there is one specific configuration key that anyone working in a production environment needs to set immediately: `attribution.sessionUrl`. By default, Claude Code has been appending these unique session tracking URLs to the automated git commits it generates.

James Turner

Wait, [scoffs] tracking links directly in the git commit messages? That is a massive red flag for enterprise compliance and basic privacy.

Lachlan Reed

Yeah, it raised a few eyebrows in the community. But thankfully, with this new update, you can scrub those links entirely. You just run `/config attribution.sessionUrl=false` and it completely strips those tracking links out of your automated git history.

James Turner

That `attribution.sessionUrl` override is going to be standard in every developer's global gitconfig by tomorrow morning. If you are signing your commits cryptographically, you absolutely cannot have external telemetry URLs polluting your immutable log.