CC Session Reader

cc-session-reader

English | 繁體中文

A CLI tool that reads Claude Code session transcripts and emits a compact summary.
Every tool call is collapsed into a single line (tool name + key arguments + result status), while conversation text is preserved in full.
Purely static extraction — no LLM involved.

Token reduction depends on what the session is made of: tool-I/O-heavy sessions typically reach 80–88%;
sessions dominated by large plan documents or plain conversation reduce less (measured around 40–65%), because user/assistant text is kept verbatim and never compressed.

Installation

macOS / Linux

The script downloads the binary for your platform into ~/.local/bin/cc-session (override with the INSTALL_DIR environment variable)
and installs the Claude Code Skill by default. Pass --no-skill if you don't want the Skill.

Windows (PowerShell)

Installs into $env:LOCALAPPDATA\cc-session\; in interactive mode it asks whether to add the directory to PATH and install the Skill.

Other installation methods

- Download a binary: grab the archive for your platform from GitHub Releases, extract it, and put it on your PATH.
- go install: go install github.com/Mapleeeeeeeeeee/cc-session-reader/cmd/cc-session@latest (binary lands in $GOPATH/bin).
- Skill only: mkdir -p ~/.claude/skills/cc-session && curl -o ~/.claude/skills/cc-session/SKILL.md https://raw.githubusercontent.com/Mapleeeeeeeeeee/cc-session-reader/main/SKILL.md

Subcommands

Session IDs support prefix matching — the first 8 characters are usually enough. read and context truncate at 200 lines by default (-max-lines adjusts it); when truncated they print the total line count and the suggested offset for the next chunk.

list reads from session metadata (~/.claude/usage-data/session-meta/), so it shows fewer entries than…