=============== LIBRARY RULES =============== From library maintainers: - Needs Cursor desktop running locally (Cloud Agents have no claude CLI), Claude Code installed and signed in, and a POSIX shell on macOS or Linux. Runs bill the Claude Code account. ### Install via Claude Code Plugin Command Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/README.md Installs the plugin in Claude Code. This does not replace the Cursor install. ```text /plugin install cursor-subagent-models --marketplace ysskrishna/claude-code-subagents-for-cursor ``` -------------------------------- ### List available skills with npx skills add --list Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/install/skills-sh-setup.md Lists what the repository offers before installing. ```bash npx skills add ysskrishna/claude-code-subagents-for-cursor --list ``` -------------------------------- ### Install skills with gh skill install Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/install/gh-cli-setup.md Installs all skills from the repository for Cursor into the user scope. Requires a GitHub CLI version that supports `gh skill`. Without `--agent cursor`, it installs for GitHub Copilot; use `--scope project` to install into the current git repository. ```bash gh skill install ysskrishna/claude-code-subagents-for-cursor --all --agent cursor --scope user ``` -------------------------------- ### Allow tools in .claude/settings.json Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/guides/permissions.md Add rules to the project's .claude/settings.json to allow specific tools for Claude Code. The example allows WebSearch and WebFetch for docs.example.com. Rule syntax is documented in Claude Code's permissions guide. ```json { "permissions": { "allow": ["WebSearch", "WebFetch(domain:docs.example.com)"] } } ``` -------------------------------- ### Cursor Plugin Installation - GitHub Repository URL Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/README.md Enter this repository URL in Cursor's plugin manager to install the skill. Choose user scope for all projects or project scope for team collaboration. ```text github.com/ysskrishna/claude-code-subagents-for-cursor ``` -------------------------------- ### Install skill with npx skills add Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/install/skills-sh-setup.md Installs the skill globally to `~/.agents/skills/` for Cursor. Drop `-g` to install into the current project instead. The repo has one skill, so no `--skill` flag is needed. ```bash npx skills add ysskrishna/claude-code-subagents-for-cursor -a cursor -g ``` -------------------------------- ### Write brief.txt with goal and search path Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/references/example.md Provides the subagent with the goal, search path, and return format. The file is written to /brief.txt. ```text Goal: find where retry settings are configured. Search: /work/api Return: up to 5 file paths, one line each, then a 2 line summary. ``` -------------------------------- ### Update and remove skills Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/install/gh-cli-setup.md Updates all installed skills and removes the specific skill directory. The `rm -rf` command deletes the installed skill folder. ```bash gh skill update ``` ```bash rm -rf ~/.cursor/skills/use-claude-code-subagents ``` -------------------------------- ### Update and remove skill with npx skills Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/docs/install/skills-sh-setup.md Updates the skill to the latest version and removes it when no longer needed. ```bash npx skills update npx skills remove use-claude-code-subagents ``` -------------------------------- ### Write agents.json with scout configuration Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/references/example.md Defines the agent configuration for the run, including description, prompt, and model. The model is the stripped version 'claude-haiku-5-5'. ```json { "scout": { "description": "Finds code and summarizes it. Read-only.", "prompt": "You find code. Return file paths and a short summary. Never edit files.", "model": "claude-haiku-5-5" } } ``` -------------------------------- ### Scout agent frontmatter and instructions Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/examples/subagents/README.md Defines a read-only subagent named 'scout' with the 'haiku' model. The frontmatter sets name, description, model, and readonly; the body contains its instructions. ```markdown --- name: scout description: Finds files and code in the project and summarizes them. Read-only. model: haiku readonly: true --- You find things in this project and report back. You never edit files. ``` -------------------------------- ### Run Claude Code subagent with agents.json Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/references/example.md Executes the scout agent with the specified model and permissions. The command uses the temp folder paths and outputs JSON. ```bash claude -p --agents "/var/folders/ab/xyz/T//claude-subagent.k3Fz9a/agents.json" --agent "scout" --model "claude-haiku-5-5" --permission-prompts none --output-format json < "/var/folders/ab/xyz/T//claude-subagent.k3Fz9a/brief.txt" ``` -------------------------------- ### Scout agent definition in .cursor/agents/scout.md Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/references/example.md Defines the scout agent with a read-only model and instructions. The model name 'claude-haiku-5-5-medium' is stripped to 'claude-haiku-5-5' for the run command. ```markdown --- name: scout description: Finds code and summarizes it. Read-only. model: claude-haiku-5-5-medium readonly: true --- You find code. Return file paths and a short summary. Never edit files. ``` -------------------------------- ### Create a temporary run folder Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/SKILL.md Creates a temporary directory outside the workspace to hold briefs and agent files. Use a new folder for each run and use the printed path literally, as the Write tool takes a literal path. ```bash mktemp -d "${TMPDIR:-/tmp}/claude-subagent.XXXXXX" ``` -------------------------------- ### Run Claude Code agent Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/SKILL.md Invokes a Claude Code agent from .claude/agents or ~/.claude/agents with the mapped model, no permission prompts, and JSON output. The brief is read from /brief.txt. ```bash claude -p --agent "" --model "" --permission-prompts none --output-format json < "/brief.txt" ``` -------------------------------- ### Cursor agent configuration JSON Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/skills/use-claude-code-subagents/SKILL.md Defines a Cursor-format agent in Claude Code format, written to /agents.json. Add --agents "/agents.json" to the claude command. Include "tools" if frontmatter has a tools list; escape quotes and newlines in the prompt. ```json { "": { "description": "", "prompt": "", "model": "" } } ``` -------------------------------- ### Activate the skill with /use-claude-code-subagents Source: https://github.com/ysskrishna/claude-code-subagents-for-cursor/blob/main/README.md Type this command in Cursor to activate the skill. After activation, the agent replies 'Subagent mode: claude code' and custom subagents run through Claude Code until you say 'use cursor subagents' to switch back. ```text /use-claude-code-subagents ``` === COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.