Agents

This repository contains local agent definitions under agents/, local skills
under skills/, and instructions under instructions/.

Instructions

# Codex
ln -sf $PWD/instructions/AGENTS.md ~/.codex/AGENTS.md

# Claude:
ln -sf $PWD/instructions/AGENTS.md ~/.claude/CLAUDE.md

# VSCode Copilot:
ln -sf $PWD/instructions/AGENTS.md ~/.copilot/copilot-instructions.md

# Cursor:
mkdir -p ~/.cursor/rules
ln -sf $PWD/instructions/cursor-instructions.mdc ~/.cursor/rules/agents.mdc

# Continue
mkdir -p ~/.continue/rules
ln -sf $PWD/instructions/continue-instructions.mdc ~/.cursor/rules/agents.mdc

Local Agents

Current local agents:

• architecture
• code-qa-engineer
• code-reviewer
• devops-engineer
• performance-engineer
• principal-engineer
• researcher-engineer
• security-engineer
• ux-engineer

Note: Keep this list in sync with the agents/ directory or generate it from the filesystem.

Link all local agents into ~/.claude/agents or ~/.cursor/rules/agents, ~/.copilot/agents is ignored since Copilot looks at claude agents and copilot agents:

mkdir -p ~/.claude/agents
mkdir -p ~/.cursor/rules/agents
for f in agents/*.agent.md; do
  ln -sf "$PWD/$f" "$HOME/.claude/agents/$(basename "$f")"
  ln -sf "$PWD/$f" "$HOME/.cursor/rules/agents/$(basename "$f")"
done

Remove agents

for f in agents/*.agent.md; do
  rm -f "$HOME/.claude/agents/$(basename "$f")"
  rm -f "$HOME/.cursor/rules/agents/$(basename "$f")"
done

External agents

• svelte-file-editor: An agent that can read, write, and edit files in a Svelte 5 codebase.

for agent in \
  https://raw.githubusercontent.com/sveltejs/ai-tools/refs/heads/main/tools/agents/svelte-file-editor.md
do
  filename=$(basename "$agent")
  curl -L "$agent" -o "$HOME/.claude/agents/$filename"
  curl -L "$agent" -o "$HOME/.cursor/rules/agents/$filename"
done

Remove external agents

for agent in \
  https://raw.githubusercontent.com/sveltejs/ai-tools/refs/heads/main/tools/agents/svelte-file-editor.md
do  filename=$(basename "$agent")
  rm -f "$HOME/.claude/agents/$filename"
  rm -f "$HOME/.cursor/rules/agents/$filename"
done

Local Skills

Current local skills:

• change-review: Provides a structured code review of a proposed change, including feedback on correctness, style, maintainability, and potential impacts.
• github-actions: Provides best practices and guidance for GitHub Actions workflow design, review, and troubleshooting. Use when you need correct, secure, and maintainable CI/CD workflow recommendations.
• openai-reusable-prompts: Designs and improves prompt best practices and formatting structure for OpenAI use cases. Use when you need clear, consistent prompt templates, stronger instruction writing, and better formatted prompt sections.
• poc: Produces a focused proof-of-concept plan, feasibility assessment, and minimal experiment artifacts for a feature, integration, or architectural approach.
• refactor: Refactor code to improve its structure, readability, and maintainability without changing its external behavior.
• review: Review or analyze something, such as a document, code, or data, to provide feedback, identify issues, or suggest improvements.
• rewrite: Performs a structured, opinionated, and safety-conscious full rewrite that intentionally introduces breaking changes.
• simplify: Simplify and unify code by improving composability, clarifying module boundaries, and producing small, easy-to-use APIs.

Install all local skills:

for d in skills/*/; do
  npx skills add "./${d%/}" --global --symlink -y
done

Install a single local skill directly from the Github repo without cloning it first:

npx skills add nathanfaucett/agents/skills/change-review --global --symlink -y

Clean up local skill links:

for d in skills/*/; do
  npx skills remove "${d%/}" --global -y
done

Optional External Skills

• caveman: Enable compressed, token-efficient assistant responses.
• grill-me: Deeply interview the user to stress-test a plan or design.
• improve-codebase-architecture: Identify architecture improvements for testability and AI navigation.
• to-issues: Break a PRD into independent implementation issues.
• to-prd: Break a high-level goal into a structured PRD with features, user stories, and acceptance criteria.
• tdd: Apply red-green-refactor workflows.
• ubiquitous-language: Establish a common language across the team to improve communication and understanding.
• svelte-code-writer: A skill that can write Svelte code based on user instructions, including components, stores, and more.
• svelte-core-bestpractices: A skill that provides best practices and guidance for writing Svelte code, including component design, state management, and performance optimization.
• web-design-guidelines: Follow best practices for web design and user experience.

for skill in \
  mattpocock/skills/skills/engineering/improve-codebase-architecture \
  mattpocock/skills/skills/engineering/to-issues \
  mattpocock/skills/skills/engineering/to-prd \
  mattpocock/skills/skills/engineering/tdd \
  sveltejs/ai-tools/tools/skills/svelte-code-writer \
  sveltejs/ai-tools/tools/skills/svelte-core-bestpractices \
  vercel-labs/agent-skills/skills/web-design-guidelines
do
  npx skills add "$skill" --global --symlink -y
done

MCP

• @sveltejs/mcp