March 31, 2026
Terminal pets, terminal opinions
Claude Code Unpacked : A visual guide
Terminal pets, 500k-line shock, and a comment war between true believers and eye-rollers
TLDR: A flashy guide to Claude Code’s unshipped features—like a terminal pet, memory between sessions, and phone control—lit up the comments. Fans hype the future; skeptics blast 500k-line bloat and pretty but empty demos. It matters because it hints at AI tools getting stickier, smarter—and more controversial.
Claude Code just dropped a glossy visual guide to its hidden features—think a virtual pet in your terminal, long-thinking sessions that can run for 30 minutes, and an assistant that remembers what happened between chats and can even be controlled from your phone. The vibe: sci‑fi wishlist meets “coming soon” tease. Fans are enchanted; one user begged for a “Claude Code spirit animal,” while others swooned over sessions that can talk to each other. You can peek at the shiny promises in Claude Code Unpacked, but the real fireworks are in the comments.
Cue the skeptics. A sharp-tongued critic said they’d use it for class as an example of what not to do, joking about “BashTool.ts,” and delivering the line of the day: “the emperor has no clothes.” Another went full disbelief with “How the hell is it 500k lines?”—sparking a mini roast about bloat. One crowd grumbled that the slick animations are all vibes, no substance, claiming it explains little beyond “you ask, it does a thing.” Meanwhile, a surreal comment—“cool Archaeologization Collection Output”—got meme’d into an in-joke about digging up ancient promises.
Bottom line: half the room is pet-naming and dreaming of autonomous helpers; the other half is calling it overproduced smoke and mirrors. Bold roadmap or bloated hype? The comments are the real showdown, and nobody’s holding back.
Key Points
- •The guide lists hidden, unshipped features in the Claude Code codebase, accessible via feature flags or environment gates.
- •A persistent mode is planned, including daily logs, memory consolidation across sessions, and autonomous background actions.
- •Long planning sessions on Opus-class models are described, with execution windows up to 30 minutes.
- •Multi-agent orchestration is outlined: a lead agent splits tasks and spawns parallel workers in isolated git worktrees, then aggregates results.
- •Remote control from phone or browser with permission approvals and cross-session communication via Unix domain sockets are planned; a terminal virtual pet is also mentioned.