Review Skills

---

description: Run skill quality review with repo-specific checks for claude-code-skills allowed-tools: Skill, Bash, Grep, Glob, Read, AskUserQuestion

Review Skills

Universal review (ln-162) + repo-specific checks for claude-code-skills repository.

Execution Strategy

**Step 1 FIRST** — invoke ln-162 via Skill tool (MANDATORY, visible to user). **Step 2** — run repo-specific bash script (can overlap with ln-162 phases). Combine in Step 3.

Step 1: Universal Review (MANDATORY Skill invocation)

**NON-NEGOTIABLE:** You MUST call `Skill(skill: "ln-162-skill-reviewer", args: "$ARGUMENTS")` via the Skill tool. Do NOT skip this or execute ln-162 phases manually without the Skill call. The Skill tool loads the full skill content which you then follow.

Invoke `Skill("ln-162-skill-reviewer")` with `$ARGUMENTS` (pass through scope — skill dirs or empty for auto-detect).

Step 2: Repo-Specific Checks

Run this single combined script. It outputs a ready-to-use report table.

#!/usr/bin/env bash

RESULTS=()
add_result() { RESULTS+=("$1|$2|$3"); }

# === R1: Marketplace paths ===
if [ -f .claude-plugin/marketplace.json ]; then
  R1_FAILS=0
  while read -r path; do
    [ -d "$path" ] || R1_FAILS=$((R1_FAILS + 1))
  done < <(grep -oE '"\./ln-[^"]+' .claude-plugin/marketplace.json | tr -d '"')
  [ "$R1_FAILS" -eq 0 ] && add_result R1 "Marketplace paths" PASS || add_result R1 "Marketplace paths" "FAIL ($R1_FAILS missing dirs)"
else
  add_result R1 "Marketplace paths" SKIP
fi

# === R2: Root docs stale skill names ===
R2_FAILS=0
for doc in README.md AGENTS.md .claude-plugin/marketplace.json; do
  [ -f "$doc" ] || continue
  while read -r skill; do
    ls -d ${skill}*/ >/dev/null 2>&1 || R2_FAILS=$((R2_FAILS + 1))
  done < <(grep -oE 'ln-[0-9]+-[a-z-]+' "$doc" | sort -u)
done
[ "$R2_FAILS" -eq 0 ] && add_result R2 "Root docs stale names" PASS || add_result R2 "Root docs stale names" "FAIL ($R2_FAILS stale refs)"

# === R3: Skill count accuracy ===
actual=$(ls -d ln-*/SKILL.md 2>/dev/null | wc -l)
R3_FAILS=0
if [ -f README.md ]; then
  badge=$(grep -oE 'skills-[0-9]+' README.md | grep -oE '[0-9]+' || true)
  [ -n "$badge" ] && [ "$badge" != "$actual" ] && R3_FAILS=$((R3_FAILS + 1))
fi
if [ -f .claude-plugin/marketplace.json ]; then
  market=$(grep -oE '"\./ln-[^"]+' .claude-plugin/marketplace.json | wc -l)
  [ "$market" != "$actual" ] && R3_FAILS=$((R3_FAILS + 1))
fi
[ "$R3_FAILS" -eq 0 ] && add_result R3 "Skill count accuracy" "PASS ($actual skills)" || add_result R3 "Skill count accuracy" "FAIL (badge/marketplace mismatch, actual=$actual)"

# === R4: Plugin completeness ===
R4_FAILS=0
for skill_dir in ln-*/; do
  skill_name="./${skill_dir%/}"
  grep -q "\"$skill_name\"" .claude-plugin/marketplace.json 2>/dev/null || R4_FAILS=$((R4_FAILS + 1))
done
[ "$R4_FAILS" -eq 0 ] && add_result R4 "Plugin completeness" PASS || add_result R4 "Plugin completeness" "FAIL ($R4_FAILS orphan skills)"

# === R5: Pipeline data-flow (semi-automated) ===
R