Add cctrack to Agent infrastructure

[nvwalj]

Project: cctrack

Repo: https://github.com/nvwalj/claude-cost-tracker

What it is

Per-project cost tracker for Claude Code. Walks ~/.claude/projects/**/*.jsonl (the same NDJSON session-telemetry files Claude Code already writes), parses every assistant message.usage block, multiplies tokens (input / output / cache-read / cache-write 5m / cache-write 1h) by published Claude 4.x prices, and prints a per-project or per-day dollar breakdown.

How it slots into this list

It's not an agent itself — it's the kind of sidecar that the "Agent infrastructure" section already lists (observability/management around CLI coding agents). Closest analogues already in the list: AgentManager (lightweight CLI for managing agent runs/sessions), brood-box (cost-tracking via VM logs in some setups), claude-code-tools (workflow utilities).

Inclusion criteria

• ✅ CLI / terminal interface: cctrack, single binary script, output is a table or --json
• ✅ Open source: MIT
• ✅ Built for agents: reads Claude Code's own telemetry format; nothing else uses these files in this way
• ✅ Single-file Python, stdlib-only (no pip dependencies)

Sample output

$ cctrack --days 7
PROJECT                                COST USD    IN tok     OUT tok    CACHE-R     CACHE-W
-------------------------------------  --------    ------     -------    -------     -------
/Users/panda/Project/ai-memory-reader  $ 41.23       2,141   1,367,884   23,912,663   2,185,484
...

License

MIT

[nvwalj]

Closing — this project has been archived / made private. If you've already merged, feel free to remove the entry; otherwise no action needed. Apologies for the noise.