cchistory

Why it's ranked here

This is a focused forensic utility with a reproducible capture path and readable output. Released builds and local binaries use the same tracing pipeline, while existing snapshots are skipped to avoid repeated downloads and model requests.

What's good

It supports single releases, release ranges, and unpublished local builds. Request selection favors non-Haiku traffic containing both tools and a system prompt. Tool entries are sorted for stable reading, MCP tools are excluded, and supplied Claude arguments are parsed with shell operators removed.

Tradeoffs

Every capture requires Claude Code authentication and sends a Haiku request, which can cost money. Released-version handling depends on a deliberately fragile text patch that may break when upstream code changes. Output omits MCP tools, and malformed trace lines fail during direct JSON parsing.

How to use it well

Use it when auditing prompt or tool changes across releases, checking a fork before publication, or preserving readable evidence of runtime behavior. Keep snapshots under version control and compare them with normal diff tools. It does not replace general request tracing or provide a comparison interface.

Observed

License

MIT

Primary language

TypeScript

Install surface

Global npm package @mariozechner/cchistory

Interface

Command-line tool named cchistory

Runtime

Node.js 16 or newer

Output

Markdown snapshot files containing user message, system prompt, and non-MCP tools

Testing

Vitest scripts are configured, with colocated tests shown for patching and JSONL parsing