Claude Code Skills & Agents
Curated skills, sub-agents, and config templates that supercharge Claude Code with deep research, image gen, GitHub automation & more.
Evidence: README requires user confirmation, default mode acceptEdits with command confirmation, warns about bypassPermissions; handoff skill has redaction; dependencies like copilot-gateway require user authentication; external effects like GitHub operations need user authorization; backup and restore advice provided. Deductions: no detailed least-privilege strategy, no data flow transparency, no dependency security audit, no rollback mechanism, no source verification.
Evidence: README and skill descriptions are consistent; skill list and config templates clear; dependencies like yt-dlp, chrome-devtools-mcp need manual install without version pinning; failure messages like copilot-gateway auth expiry have FAQ guidance. Deductions: dependency availability unverified, failure messages incomplete.
Evidence: Targets diverse users (developers, researchers), scenarios broad (research, image generation, GitHub automation); trigger words explicit, config templates for multiple providers; environment requirements (Python, Node) stated. Deductions: capability boundaries not explicit, no environment compatibility tests.
Evidence: README well-structured, install instructions detailed (multiple methods), skill naming consistent, FAQ covers common issues, MIT license clear. Deductions: no version history or changelog, maintenance responsibility not explicit (only author attribution).
Evidence: Skill outputs clear (e.g., deep-research generates reports), multiple config templates reduce cost, high marginal value (integrates multiple capabilities). Deductions: no performance benchmarks or cost estimates.
Evidence: README references official docs and external resources, skill descriptions specific. Deductions: no verifiable test results, no separation of facts and inferences.
- Manual installation overwrites ~/.claude directory; backup first.
- Using bypassPermissions mode has security risks; proceed with caution.
- Dependencies like copilot-gateway require GitHub authentication; be mindful of credential security.
- Skills installed via npx skills may lag behind; prefer official marketplace.
What does this agent do, and when should you use it?
This repository offers a curated set of skills, sub-agents, and configuration templates for Claude Code, covering brainstorming, deep research, image generation (Gemini/OpenAI), YouTube transcription, translation, GitHub issue fixing and PR review. It installs via the official Claude marketplace plugin or npx skills, with skills defined as SKILL.md files that can be triggered manually or automatically. Sub-agents like pr-reviewer and github-issue-fixer are specialized agents for code review and issue fixing. Configuration templates support switching between model providers such as GitHub Copilot, LiteLLM, DeepSeek, and Qwen. A script provides one-click updates for plugins. The core depends on Claude Code, but it is adaptable to different model providers.
The repository includes 12 skills: brainstorming (progressive requirement exploration), codex-skill (delegates to Codex CLI for coding and review), deep-research (multi-agent parallel research with structured report generation), github-fix-issue (end-to-end issue fixing), github-review-pr (parallel sub-agent PR review with adversarial validation), gpt-image-skill and nanobanana-skill (image generation via OpenAI and Gemini respectively), grill-me (intensive design interrogation), handoff (generates handoff documents), skill-creator (creates and benchmarks skills), translate (translates tech articles into Chinese), and youtube-transcribe-skill (extracts subtitles). Sub-agents include pr-reviewer, github-issue-fixer, and others. Configuration templates provide pre-set settings.json for different providers, requiring ANTHROPIC_BASE_URL and API key changes. The script scripts/update-cc-plugins.sh updates plugins.
- A developer wants to add deep research capability to Claude Code; install the deep-research skill and provide a query to get a parallelized research report.
- A user needs to generate or edit images; use nanobanana-skill by default or gpt-image-skill for OpenAI, providing prompts to call Gemini or GPT Image APIs.
- A maintainer wants to automate GitHub issue fixing; use the github-fix-issue skill for the full flow from analysis to PR creation.
- A team wishes to run Claude Code with non-Claude model providers like GitHub Copilot or DeepSeek; copy the corresponding config template and start copilot-gateway or set the endpoint.
- A translator needs to translate English or Japanese tech articles into Chinese; use the translate skill which preserves Markdown formatting.
- A developer wants to create custom skills; use the skill-creator skill for iterative development and benchmarking.
What are this agent's strengths and limitations?
- Comprehensive skill collection covering broad development tasks from brainstorming to GitHub automation, ready to use out of the box.
- Ten configuration templates for switching between GitHub Copilot, DeepSeek, Qwen, and other providers, reducing vendor lock-in.
- deep-research supports multi-agent parallel research with structured output, ideal for complex investigations.
- github-review-pr incorporates adversarial validation and confidence scoring, improving review rigor.
- Depends on Claude Code and Anthropic toolchain; core skills cannot run natively on other coding assistants.
- Manual installation overwrites ~/.claude directory, requiring backup and careful configuration.
- npx skills installs may lag behind the repository version, requiring version synchronization.
- Configuring multiple model providers requires understanding environment variables like ANTHROPIC_BASE_URL, with a learning curve for new users.
- Some skills (e.g., gpt-image-skill) need extra API keys and Python environment setup.
How do you install or deploy this agent?
In Claude Code session, use slash commands:
/plugin marketplace add feiskyer/claude-code-settings
/plugin install claude-code-settingsOr use CLI:
claude plugin marketplace add feiskyer/claude-code-settings
claude plugin install claude-code-settings@claude-code-settings
For individual skills: npx -y skills add --all feiskyer/claude-code-settingsNote: ~/.claude/settings.json must be configured manually.
How do you use this agent?
After installation, skills are invoked via /skill-name or triggered automatically by context. For example, type '/deep-research your topic' to use deep-research. To switch model providers, copy the appropriate template (e.g., settings/copilot-settings.json) to ~/.claude/settings.json and set ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY to your gateway endpoint. For GitHub Copilot, first run 'npx copilot-gateway@latest start --proxy-env' and complete device authentication.
How does this agent compare with similar options?
The official anthropics/skills repository provides similar but more basic skills; this repository integrates more advanced skills and sub-agents.