Claudeview — Design skill for Claude Code

**What is it?**

claudeview is a zero-setup TUI that gives you full observability into Claude Code sessions. It reads `~/.claude/` directly — no hooks, no config, no agents to run. Projects, sessions, agents, tool calls, plugins, and memories are all navigable from a single terminal interface with live-streaming updates.

**Core Design**

The dashboard follows the k9s model: hierarchical drill-down with vim-style navigation. `j/k` to move, `enter` to drill down, `esc` to go back, `/` to filter. Every view refreshes on a 1-second tick, and the history view supports follow mode — auto-scrolling to the latest turn like `tail -f` as Claude writes.

**Key Capabilities**

1. Session hierarchy — projects, sessions, agents, and tool calls in a single navigable tree
2. Slug grouping — related sessions (plan/execute transitions sharing a slug) collapse into a single row with aggregated stats and a merged history view separated by divider rows
3. Subagent visibility — full agent tree with types, status, and embedded transcript rendering in the history view
4. Live follow mode — history view streams new turns as they arrive; scroll up to lock position, G to re-enable
5. Plugin and memory inspection — jump to plugins (p) or memories (m) from any view, with state preserved for esc-to-restore
6. Detail views — expanded content with thinking blocks, tool call inputs/outputs, and per-model token counts
7. Usage monitor — █░ progress bars for 5-hour and 7-day Claude Max token windows with reset countdowns, shown above the info panel; hidden automatically when no credentials are found

**Getting Started**

Install the latest release:

curl -fsSL https://raw.githubusercontent.com/Curt-Park/claud