claude-jev
claudeFive hooks, one batched API call each. Classifies every prompt into a one-line routing hint; sets the model tier a subagent spawns on; judges each edit against the rules already in your CLAUDE.md, AGENTS.md and .claude/rules, citing the one it breaks; and on compaction judges 150 transcript blocks keep, truncate or drop so the survivors carry over verbatim instead of through a generated summary. Fails open everywhere. Shaped by evals over 2,464 real prompts and 250 real edits; see eval/.
4 skills · hooks