KKitforgeinkitforge.hashnode.dev·Sep 13 · 3 min readThe AI Code Review Checklist: What to Actually Check in Agent-Written CodeReviewing agent-written code with a human-code checklist misses the ways agents actually fail. Agent output rarely has typos or sloppy formatting - it is fluent, plausible, and wrong in specific, repe00
KKitforgeinkitforge.hashnode.dev·Sep 13 · 4 min readClaude Code Keeps Forgetting Your Project? How CLAUDE.md Survives the Context WindowAn hour into a session, Claude Code stops following the conventions you explained at the start. It renames things you told it not to rename, forgets which service owns the database, proposes the patte00
KKitforgeinkitforge.hashnode.dev·Sep 12 · 4 min readClaude Code Hooks Not Firing? The 6 Causes That Cover Almost Every CaseYou wrote the hook, saved the settings, ran the tool - and nothing happened. Hooks fail silently by design: a broken hook should not block your work, so errors get logged instead of raised. That makes00
KKitforgeinkitforge.hashnode.dev·Sep 12 · 3 min readAGENTS.md vs CLAUDE.md vs .cursorrules - Which Agent Config File Do You Need?Three files, three ecosystems, one job: tell the AI coding agent how your repo works before it starts guessing. They overlap enough that teams routinely cargo-cult all three with contradictory content00
KKitforgeinkitforge.hashnode.dev·Sep 12 · 3 min readCLAUDE.md in a Monorepo - Nested Files, Imports, and What Actually WorksA single root CLAUDE.md works until the repo has twelve packages with different stacks, test commands, and conventions. Then the file either balloons into a 900-line document that costs tokens on ever00