Source: ai-research/mulerun-skillshub-2026-05-28.md

URL: https://mulerun.com/skillshub Featured skill categories surfaced in extract: skill maintenance (skill-creator/updater/validator, plugin-validator, link-checker), code execution (code-runner), media (generating-memes, video-comparer), text processing (transcript-fixer, figlet-text-converter), session/history recovery, competitive analysis.

A second third-party Claude/Codex skill-discovery directory, sister to claudeskills.info but with two structural differences: it surfaces per-skill RUN counts (raw execution tallies, not stars/installs) and it foregrounds skill-maintenance utilities (skill-creator, skill-updater, skill-validator, plugin-validator, link-checker) as the lead catalog tier. Skills are authored for both .claude/skills/ and .codex/skills/ directories, with AGENTS.md awareness baked into the lead skill-creator skill.

Key Takeaways

  • Dual-runtime by default. The lead skill-creator skill targets both Claude Code (.claude/skills/) and Codex (.codex/skills/) directory conventions in one entry — the Hub doesn’t fork the catalog by runtime.
  • RUN counts as engagement metric. Each card displays a numeric RUN count rather than stars or installs (observed values: 15, 16, 602). Interpret as raw execution counts — community signal, not a quality bar.
  • Skill-maintenance tier leads the catalog. First-class billing for skill-creator / skill-updater / skill-validator / plugin-validator / link-checker positions the Hub as a workshop layer for skill authors, not only a discovery layer for consumers.
  • AGENTS.md-aware skill generation. The skill-creator description explicitly mentions aligning generated skills against AGENTS.md rules — directly compatible with the per-topic AGENTS.md convention in the karpathy vault schema.
  • Differentiates from claudeskills.info on two axes. claudeskills.info surfaces cross-vendor publisher collections (Anthropic, OpenAI, Microsoft, Google, Vercel, GitHub Copilot, WordPress) with stars/installs; MuleRun surfaces RUN-count engagement on individual maintenance and utility skills. Use both for complementary discovery, not redundant lookup.
  • Trust posture: discovery surface, not a vetting authority. ^[inferred] No publisher tiers or curator vetting surfaced in the extract — treat individual skills with the same caution as raw GitHub finds. Use Hub → underlying repo → manual review before installing.
  • Extract was partial. Tavily’s extract_depth: advanced returned individual skill cards but truncated the category navigation, install instructions, and counts beyond the visible RUN tallies. Full taxonomy enumeration would require a second pass.

Notable cataloged skills

SkillPurposeRUN
skill-creator / skill-updater / skill-validatorCreate / update / validate skills against codebase + AGENTS.md rules; tune YAML triggers; generate <available_skills> blocks16
code-runnerExecute code in 30+ languages16
generating-memesMeme CLI with 298 templates16
plugin-validatorValidate plugin structure, manifests, silent-failure patterns15
figlet-text-converterReplace figlet tags with ASCII art; preserves comment styles; 400+ fonts15
link-checker / portabilityValidate links + absolute-vs-relative paths in skills/plugins15
session-history recoveryRecover deleted files / find work from previous .claude/projects/ sessions
competitors-analysisEvidence-based competitor tracking (code-grounded, not assumption-based)602
transcript-fixerCorrect ASR/STT errors, homophones, mixed-language transcripts; learns personalized correction dictionaries602
video-comparer(description truncated in extract)

Try It

  1. Visit https://mulerun.com/skillshub to browse the live catalog.
  2. To compare against the wiki’s existing reference directory, cross-check with claudeskills.info — note which skills appear in one but not the other (RUN-count framing tends to surface different leaders than star-count framing).
  3. To install a candidate skill: follow the link from the Hub card to the underlying GitHub repo, audit the SKILL.md against your project’s needs, and install via the standard Claude Code skills path (.claude/skills/<name>/) — the Hub does not appear to ship its own one-click installer in the extracted snapshot.
  4. For skill authors: the lead skill-creator tool is itself a candidate skill to install — it would automate the create/update/validate cycle of skills you write, aligned to your repo’s AGENTS.md.

Open Questions

  • Full category taxonomy — extract truncated category navigation; a second pass (browser-driven scrape or page-level Tavily walk) would enumerate the full structure.
  • Install/submission flow — does the Hub support PR-based submission, one-click install, or pure discovery + link-out only?
  • RUN counter source — is the RUN count Hub-internal telemetry, self-reported by skill authors, or aggregated from a CLI? Affects how to interpret 602 vs 15 as quality signal.
  • Overlap with claudeskills.info / SkillsMP — how much of the catalog is the same canonical set vs unique to MuleRun? A diff would inform whether to track both or pick one as canonical.