Session Companion — AI skill for Claude Code

Session Companion

**Unofficial, community-built.** Not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" and "Claude Code" are Anthropic's trademarks, referenced here only to describe compatibility.

A VS Code extension that reimplements the official Claude Code chat UI on top of [`@anthropic-ai/claude-agent-sdk`](https://www.npmjs.com/package/@anthropic-ai/claude-agent-sdk), with one addition the official extension doesn't have: a **Sessions** view that aggregates Claude Code session history across every repo on the machine, not just the current workspace.

Features

• Streaming chat, token-by-token, with markdown rendering and collapsible step cards — tool calls, extended-thinking blocks, and auto-injected IDE context notes (e.g. "opened file X") each collapse to a one-line summary once finished, so a response with many steps stays readable instead of dumping everything open. A Task tool call nests its subagent's own text/thinking/tool-use as flat lines under the same card (Options.forwardSubagentText), with a live present-tense progress summary while it's running (Options.agentProgressSummaries)
• Each response ends with a small stats line — duration, cost, and input/output tokens for that turn (SDKResultMessage), the same kind of info the terminal's /cost gives you, but inline per message instead of on demand. Live turns only — reopening a past session from the Sessions list doesn't retroactively show stats for its earlier turns. Investigated backfilling this from the stored session file and deliberately didn't: cost isn't persisted anywhere in it, and reconstructing tokens by summing each stored assistant message's own usage undercounted badly against the real live total for an equivalent turn (14 vs. 197 output tokens in one real test) — the stored transcript doesn't expose everything that counts toward a turn's real usage. Only duration is honestly reconstructable (via each message's un