agent-skills

A Claude Code plugin marketplace with useful skills for AI agents.

Installation

First, add the marketplace:

/plugin marketplace add mrshu/agent-skills

Then install plugins:

/plugin install scholar-search
/plugin install wshot
/plugin install paper-siphon
/plugin install codex-exec
/plugin install claude-exec
/plugin install review-anvil
/plugin install overleaf-comment
/plugin install overleaf-compile
/plugin install gdocs-comment
/plugin install gh-pr-image

For hosts that consume skills directly via npx skills, install the Review Anvil family with its reviewer backends in one command:

npx skills add mrshu/agent-skills \
  --skill review-anvil \
  --skill review-anvil-readonly \
  --skill review-anvil-pr \
  --skill review-anvil-improve-pr \
  --skill codex-exec \
  --skill claude-exec

Plugins

scholar-search

Find academic papers, explore citations, and export BibTeX. Uses four tools under the hood:

• s2cli (Semantic Scholar) — paper recommendations, citation/reference networks, arXiv/DOI lookups
• openalexcli (OpenAlex) — broadest coverage (260M+ works), institution/journal queries, aggregation
• arxivy (arXiv) — direct arXiv search, paper lookup by ID, browse recent submissions by category
• dblpcli (DBLP) — high-quality, curated BibTeX export

All tools run via uvx — no installation needed.

paper-siphon

Convert academic PDFs to clean Markdown. Supports local files and remote URLs (e.g. arXiv). Three extraction modes:

• Default — fast Docling-based extraction, handles most papers
• VLM (--vlm) — vision-language model for complex layouts and scans
• Formula enrichment (--enrich-formula) — better LaTeX fidelity for math-heavy papers

Runs via uvx — no installation needed.

codex-exec

Delegate code review, plan review, and deep exploration tasks to Codex CLI from Claude Code. Use it as a pre-push quality gate, plan counter-review, or deep-dig exploration tool. Key commands:

• codex review --base main — review current branch against main
• codex review --uncommitted — review work-in-progress changes
• codex exec '' — custom prompted review, plan critique, or codebase exploration

claude-exec

Delegate code review, plan review, and deep exploration tasks to Claude Code CLI from Codex or other agents. The mirror of codex-exec — lets non-Claude agents get an independent second opinion. Key commands:

• claude -p '' — non-interactive review or exploration
• claude -p --permission-mode dontAsk --tools "Bash,Read,Glob,Grep" ... — scripted review with an explicit built-in tool set
• See plugins/claude-exec/skills/claude-exec/SKILL.md for the full fallback recipe, including --allowedTools, stdin prompts, and watchdog use.

review-anvil

review-anvil turns a code review into a repeatable loop: codex-exec