30 秒でわかる
Theo describes features he likes after using Claude Code extensively. He wants other agent tools to adopt similar ideas, and explicitly sets aside his criticisms for this short discussion.
- 1Scripts in skillsTheo likes being able to execute scripts as part of loading a skill.
- 2Import context filesHe demonstrates importing AGENTS.md into CLAUDE.md with the @ syntax.
- 3Side questions and workflowsHe highlights a side-question command and workflows that let agents write code to coordinate work.
- 4Remote controlHe tries controlling a local session through the web or a phone app.
ポイント
- Scripts in skills can supply information when a skill loads.
- CLAUDE.md can import other files, including AGENTS.md, into its context.
- A side-question command can answer a quick question without interrupting the main conversation.
- Theo values workflows that generate code and coordinate parallel agents.
- Remote control can connect a local Claude Code session to a phone or browser.
チャプターごとに整えた全文を読む
Features worth borrowing
Theo explains that he wants other tools to adopt the features he likes. He begins with scripts in skills and context imports.
I've been using Claude Code a lot, and despite my criticisms, I've found features I really like. My goal in sharing them is for other agent harnesses to borrow the good ideas, so I don't miss them when I use something else.
One useful feature is the ability to execute scripts in skills. There are times when it is useful for the model to have information as the skill loads instead of loading the skill and then taking several separate steps to obtain it.
I have concerns about CLAUDE.md and AGENTS.md being competing conventions, but CLAUDE.md can import additional files using @path/to/import. The imported content loads alongside it. For example, referencing @AGENTS.md lets me reuse that file rather than maintain a separate set of instructions.
Side questions, workflows and remote control
He highlights a way to ask quick questions and describes agent workflows. He also tries remote control for a local session.
I also found links using the claude-cli:// scheme that can launch sessions, and I like the side-question command for asking something quickly without interrupting the main conversation. Both offer ways to access the agent without disturbing the flow of the current task.
Workflows go beyond a list of tool calls: the agent can write code to carry out a task. Theo describes the workflow he is using as running up to eight agents in parallel and starting more as work completes. He sees this as a useful approach to coordinating work.
I also tried remote control, which lets me access the Claude Code session running on my computer through the web or my phone. There are plenty of things the tool does well. I've deliberately focused on those here, though I may discuss the problems in a future video.
同じチャンネルの記事
38:47EN
Theo - t3․gg· 2026年10月28日Theo’s Long-Running Agent Experiments: Verification, Compute and CostTheo Browne · Ping LabsA skateboard game and two compiler projects show how repeated checks and flexible compute can sustain agent work—and how expensive those loops can become.
36:07ENTheo - t3․gg· 2026年10月27日Theo on Token Costs, Coding Agents, and the Price of AI CapabilityTheo Browne · Ping LabsA viral estimate of Lauren's AI spending prompts a closer look at cache pricing, benchmarks, and the changing work of coding agents.
30:37ENTheo - t3․gg· 2026年10月26日Theo’s Three Skills for Reviewing and Shipping with Coding AgentsTheo Browne · Ping LabsHow sol-loop, full-send and full-stop encode review, verification and different levels of merge authority.
0:43ENTheo - t3․gg· 2026年10月23日Theo Distinguishes Security Vulnerabilities from Active AttacksTheo Browne · Ping LabsA short explanation of why disclosed software flaws and reported attacks are different kinds of security news.
29:22ENTheo - t3․gg· 2026年10月22日Theo on Security Vulnerabilities, Stolen API Keys and AITheo Browne · Ping LabsA look at the gap between vulnerability disclosure and patching, and a speculative account of why stolen inference keys may be valuable.
31:13ENTheo - t3․gg· 2026年10月20日Theo on Why Smarter Models Help More Than DevelopersTheo Browne · Ping LabsFrom game modifications to compiler experiments, Theo revisits his assumptions about who benefits from more capable AI.