← ClaudeAtlas

verify-narulisted

Drive Naru's command-line curation, injection, recall, and maintenance flows with an isolated SQLite store. Use after changes to naru.py, ms.py, CLI storage, or user-visible command behavior.
9710willy/naru · ★ 0 · AI & Automation · score 70
Install: claude install-skill 9710willy/naru
# Verify Naru Naru's main user surface is the short-lived CLI in `naru.py`. The Codex hooks, Claude Code spill hook, model agent, and benchmark are other surfaces. This skill covers the CLI. Run their module self-checks when a change touches them. ## Launch Run from the repository root. This creates one state folder and one separate evidence folder. The state folder includes its own SQLite database, metrics file, and temp folder. Two runs can work side by side when each uses this block. ```bash set -eu VERIFY_NARU_HOST_TMP="${TMPDIR:-/tmp}" VERIFY_NARU_STATE=$(mktemp -d "${VERIFY_NARU_HOST_TMP%/}/verify-naru-state.XXXXXX") VERIFY_NARU_EVIDENCE=$(mktemp -d "${VERIFY_NARU_HOST_TMP%/}/verify-naru-evidence.XXXXXX") mkdir -p "$VERIFY_NARU_STATE/tmp" export VERIFY_NARU_HOST_TMP VERIFY_NARU_STATE VERIFY_NARU_EVIDENCE export NARU_DB="$VERIFY_NARU_STATE/log.db" export NARU_NOTES="$NARU_DB" export NARU_METRICS="$VERIFY_NARU_STATE/metrics.jsonl" export TMPDIR="$VERIFY_NARU_STATE/tmp" ./.agents/skills/verify-naru/capture.py launch -- python3 naru.py --help rg -q "naru claim" "$VERIFY_NARU_EVIDENCE/launch.terminal.txt" rg -q "naru inject" "$VERIFY_NARU_EVIDENCE/launch.terminal.txt" printf 'ready - state=%s evidence=%s\n' "$VERIFY_NARU_STATE" "$VERIFY_NARU_EVIDENCE" ``` The CLI is ready when the block exits `0` and prints `ready`. There is no server or long-running process. Use the Cleanup block when the drive ends. ## Doctor Run this read-only check before the first drive and when