Software engineer. I build an internal AI agent platform at a bank — mostly the unglamorous half of agents: context files, tool governance, evaluation, PII boundaries, and enough observability to prove any of it works.
🔎 driftlint — your CLAUDE.md is lying to your agent
A zero-dependency linter for agent context files. Code moves; CLAUDE.md doesn't — and unlike a human reader, the agent can't tell. driftlint finds the claims your code no longer supports: dead file references (with did-you-mean fixes), removed npm scripts and make targets, skills that overflow their listing budget and go silently invisible, config pointing at scripts that aren't there, and CLAUDE.md ↔ AGENTS.md copies that have drifted apart.
npx @alifurkangokce/driftlint # lint this repo
npx @alifurkangokce/driftlint twins # keep CLAUDE.md and AGENTS.md in syncI scanned 118 popular public repositories to check the problem was real: 59% carry at least one hard drift error.
A monthly series on AI engineering, in Turkish — prompt injection, agent observability, agent memory, LLM cost engineering, guardrails: medium.com/@alifurkangokce