ContextLens
ContextLens measures whether changes to AGENTS.md, CLAUDE.md,
skills, MCP tools, and other repository context actually improve coding-agent
performance, or just consume more tokens.
Motivation
Teams now commit agent configuration next to code, but rarely test it. A smaller prompt is not automatically a cheaper run. An agent can compensate by searching more, taking more turns, or missing the cache. ContextLens keeps repository footprint and effective task context as separate measurements.
How it works
- Static scan finds repeated guidance, missing paths, and token bloat with no API key
- Optional A/B verify replays real agent tasks against a base and a candidate
- Reports repository footprint, findings, and (when verified) agent economics