feat(ci): add check-skill-tiers lint to surface untagged skills

[naimkatiman]

Summary

Closes the silent "## Other skills" gap from the tier-rendering generator added in feat(skills): tier in frontmatter + generator-rendered tier table (already on main). The bundled README groups skills by tier; an untagged skill would slide into a catch-all bucket without anyone noticing in code review. This lint fails the build when any source skill (SKILL.md or skills/<name>.md) is missing or has an unrecognized tier: value.

The lint caught a real problem on first run: the root SKILL.md was the only source that hadn't been tagged. Added tier: core in the same commit.

What changed

• src/bin/check-skill-tiers.mts (new) — pure checkSkillTiers(repoRoot) helper for unit tests, plus a CLI main() that exits 0/1 with a clear failure message. Pattern-matches src/bin/check-skill-mirror.mts.
• src/test/check-skill-tiers.test.mts (new) — 10 tests: missing tier, unrecognized tier, ignored README.md, missing skills/ dir, variant-spelling acceptance (FEATURED, tier1, tier-2), and CLI exit-code behavior on synthetic fixtures plus a live-repo no-drift assertion.
• package.json — adds verify:skill-tiers npm script alongside the existing verify:skill-mirror and verify:docs-substrings lints.
• SKILL.md — adds tier: core (caught by the lint).
• Generated mirror + .mjs siblings.

Test plan

• npm test passes (228 → 238 tests; +10 from the new suite)
• npm run verify:skill-tiers exits 0: OK skill-tiers: all 12 skill source(s) declare a recognized tier.
• npm run verify:skill-mirror still clean (mirror updated)
• CLI surfaces clearly when a skill is intentionally untagged (manual: temporarily strip tier: from any skill, expect non-zero exit)

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces a new linting utility, check-skill-tiers, to ensure that all skill markdown files, including the core SKILL.md, declare a recognized tier in their YAML frontmatter. The implementation includes the script in both TypeScript and JavaScript, comprehensive unit and integration tests, and a new npm script for verification. The review feedback suggests improving the robustness of the core skill file check by verifying it is a file rather than a directory and optimizing the directory traversal in discoverSkillSources by using withFileTypes: true to reduce system calls.

[gemini-code-assist]

The statSync call for the core skill doesn't verify that the path is actually a file. If SKILL.md were a directory, statSync would succeed, but the subsequent readFileSync in checkSkillTiers would throw an EISDIR error and crash the script.

  try {
    const stat = statSync(corePath);
    if (stat.isFile()) {
      out.push({ name: CORE_SKILL_NAME, path: corePath });
    }
  } catch {

[gemini-code-assist]

Using readdirSync with { withFileTypes: true } is more efficient as it avoids calling statSync for every file in the loop. This reduces the number of system calls.

  try {
    const entries = readdirSync(skillsDir, { withFileTypes: true });
    for (const entry of entries) {
      if (!entry.isFile() || !entry.name.endsWith(".md") || entry.name === "README.md") continue;
      out.push({
        name: entry.name.replace(/\.md$/, ""),
        path: join(skillsDir, entry.name),
      });
    }
  } catch {
    return out;
  }