Why Claude Code Dropped Todos and Slash Commands

Over two days, Anthropic changed two things in Claude Code simultaneously. Todos became Tasks. Slash Commands became Skills. Both were framed as “existing features, same behavior” - easy to overlook. But the two changes point in exactly the same direction.

Why TodoWrite Disappeared

Anthropic’s explanation is revealing: “Opus 4.5 runs autonomously for longer and tracks state more effectively. For small tasks, TodoWrite is no longer necessary.”

• The model already figures out what needs to be done on its own
• A separate to-do list becomes overhead rather than help
• The core move is unhobbling - removing constraints that limit what the model can do

The model was already capable. The tool was holding it back.

Why Slash Commands Became Skills

The same logic applies here. Slash Commands were designed around progressive disclosure - load context only when needed. As the model got smarter, that approach stopped being optimal.

• Skills read related files automatically to assemble context
• A SKILL.MD file can reference other files, enabling multi-step context chains
• From the model’s perspective, there’s no reason to split this into a separate tool

The Common Thread: Removing Scaffolding for Simple Work

Here’s the pattern across both changes:

• Todos: The model handles it on its own, so the tool is unnecessary
• Slash Commands: The model loads context on its own, so the special syntax is unnecessary

Both follow the same philosophy: as model capability increases, remove the scaffolding that was compensating for its limitations. Anthropic calls this “unhobbling” - taking off the chains.

Yet Tasks and Skills Got More Powerful

Here’s the paradox. Tools for simple work disappeared, but the abstractions for complex work got significantly more sophisticated.

What Tasks added:

• Dependencies and blockers stored as metadata between tasks
• File-system-based coordination so multiple sessions and sub-agents can collaborate
• Updates in one session automatically reflected across all sessions sharing the same Task List

What Skills added:

• An agent: setting that creates a sub-agent with a skill loaded
• A context: fork option that clones the full current context into a separate sub-agent
• Fine-grained control over who can invoke a skill - user, model, or both

A New Principle for Agent Design

The principle I read from these two changes:

• Simple tasks: Hand them to the model and remove the tools
• Complex collaboration: Build structures for shared state and context isolation

An environment variable like CLAUDE_CODE_TASK_LIST_ID=groceries claude lets multiple sessions share the same Task List. Combined with Skills’ sub-agent integration for splitting context, you can run multiple workstreams concurrently. This is the basic skeleton of multi-agent collaboration that Claude Code is proposing.

As models get smarter, simple scaffolding disappears and only the collaboration layer remains.

The future of agent systems isn’t about betting on a single agent’s capability. It’s about designing how multiple agents split and merge state. That’s what will matter most going forward.