Claude Code is brilliant but messy to read. Long sessions blur into a wall of tool calls, code dumps, and scattered side-quests. Studio Mirror gives you the version of that session a human can actually follow.
It mirrors your active Claude Code chat as readable prose: the tool noise, the code blocks, the system reminders all stripped. What's left looks like a conversation.
Then it goes one step further: it splits every session into the tasks you actually worked on. While you're talking to Claude, the bundled skill tags each response with a short task slug , [T:auth], [T:padding-fix], [T:deployment]. The side rail lists every task, with message counts. Click any tag to isolate that thread.
This unlocks parallel work in a single session. Start one task, get pulled into a second, jump back to the first , the mirror keeps each thread coherent. No more scrolling through 200 messages to find where you left off on the thing you started an hour ago.
What you can do
- Read a session like a chat. No tool calls, no code dumps, no system noise.
- Switch tasks fluidly. Filter the thread to one task. Jump between tasks via the side rail.
- Quote and continue. Highlight any past message, write a comment, click Copy , the clipboard receives a structured block that pastes straight back into Claude Code with full context. Claude picks up the thread.
- Auto-focus new tasks. When Claude opens a new task slug, the mirror switches to it automatically.
- Live-follow. When you focus a different Claude Code tab, the mirror follows.
How it works (short)
Studio Mirror reads Claude Code's local transcript files. No API calls, no telemetry, no upload. It installs a small skill (task-threads) into ~/.claude/skills/ so Claude knows how to tag its own responses. The skill auto-installs on first run; opt out via studioMirror.manageSkill: false.
License
MIT. See LICENSE.