← ClaudeAtlas

kibitzer-sample-reviewlisted

Find real kibitzer PostToolUse hook firings in Claude Code session transcripts and record them as categorized, checkable false-positive/true-positive samples in tstapler/kibitzer's docs/
tstapler/kibitzer · ★ 0 · Code & Development · score 66
Install: claude install-skill tstapler/kibitzer
# Kibitzer Sample Review kibitzer (`github.com/tstapler/kibitzer`, installed via Homebrew as `tstapler/kibitzer/kibitzer`) is a Rust CLI that runs as a Claude Code `PostToolUse` hook. Its checks are whole-file (sometimes whole-repo) and not diff-aware — see `src/check.rs` and `src/hook.rs` in that repo. That one architectural fact explains most of its surprising firings: any edit, including a deletion or an edit unrelated to the actual violation, can re-surface a check failure that already existed, or exists only because a multi-step edit sequence is mid-way through. This skill is how to find a *specific real occurrence* of that (or any other kibitzer behavior worth tracking) and file it where the next investigation can find it without redoing the transcript archaeology from scratch. ## Step 1 — Find real invocations, not text matches Follow `checking-invocations.md` in `tstapler/kibitzer`'s `docs/` exactly — do not grep for the word "kibitzer" and treat hits as evidence; that conflates real hook firings with kibitzer's own source code being edited or discussed. The reliable filter is `attachment.command == "kibitzer hook"` (or `attachment.blockingError.command` for blocks) on `hook_success` / `hook_blocking_error` attachment records. ```bash for f in ~/.claude/projects/*/*.jsonl; do success=$(jq -c 'select(.type=="attachment" and .attachment.type=="hook_success") | select(.attachment.command=="kibitzer hook")' "$f" 2>/dev/null | wc -l) blocked=$(jq -c 'select(.ty