Summary
The sidecar records per-turn token, cache, latency and cost data, reads existing local transcripts, and stores metrics in a local SQLite database. It also exposes Prometheus metrics and mines repeated command sequences into installable SKILL.md files, while its optimisation recommendations currently measure rather than automatically change workflows.
Why it matters
As coding-agent usage becomes harder to budget and supervise, local cost attribution and workflow evidence provide a practical foundation for engineering governance.
Related coverage
- DeepSeek releases a plugin-oriented agent harness
- ChainDrop exposes persistence through Claude Code and VS Code configuration
- GraphQL tooling converges on LLM-generated mocks without a standard