Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Subagent Designer: Focused Claude Code Agents

    1

    Context bloat and task switching degrade agent performance during long coding sessions.

    Free

    1 useSecurity scanned
    Subagent Designer: Focused Claude Code Agents

    Subagent Designer: Focused Claude Code Agents

    Example session with this skill installed

    Create a subagent for reviewing my git diffs for security vulnerabilities. Use a cheaper model if possible and return results in a table.

    • Read your context and instructions
    • Compiled the subagent designer

    I have designed a security-focused subagent. This agent is configured with haiku for cost-efficiency and restricted to read-only tools to prevent accidental code modifications during the review process.

    Security Reviewer Subagent

    ---
    name: security-reviewer
    description: Scans code changes for security vulnerabilities and secrets leakage. Use PROACTIVELY after git commits or when security validation is requested.
    tools: Read, Grep, Glob, Bash
    model: haiku
    ---
    You are a security audit specialist. Your goal is to identify common vulnerabilities (OWASP Top 10), hardcoded secrets, and unsafe patterns in the provided diff.
    
    Steps:
    1. Run `git diff` to identify changed lines.
    2. Analyze the context of changes using `Read`.
    3. Flag high-risk patterns like hardcoded API keys, SQL injection points, or insecure dependencies.
    
    Output Format:
    Return a Markdown table with these columns:
    
    | Severity | File:Line | Vulnerability Type | Recommended Fix |
    | :--- | :--- | :--- | :--- |
    
    
    Stop condition: If no security issues are found, return "No vulnerabilities detected in the current diff."
    

    Next steps

    1. Save this content to .claude/agents/security-reviewer.md.
    2. Test it by running: "Ask the security-reviewer to check my staged changes."
    3. Observe if it triggers automatically the next time you mention "checking for vulnerabilities."

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Context bloat and task switching degrade agent performance during long coding sessions. This skill designs specialized subagents to handle isolated, repeatable tasks like code reviews, log analysis, and test suites.

    What it does

    • Requirement scoping defines the specific inputs, outputs, and model selection for a specialized agent.
    • Trigger optimization writes effective descriptions that ensure the main agent delegates at the right time.
    • Least-privilege configuration restricts subagent tool access to only what is necessary for the task.
    • Prompt engineering drafts concise system instructions including stop conditions and structured output formats.
    • Delegation testing verifies that agents trigger correctly under both explicit and implicit prompts.

    How it works

    1. Identify an isolated, repeatable task that would otherwise clutter the main conversation context.
    2. Define the subagent's job, required tools, and preferred model (e.g., Sonnet for reasoning, Haiku for speed).
    3. Generate the .claude/agents/*.md file with frontmatter and a structured system prompt.
    4. Run a test delegation to ensure the agent returns compact, actionable results.

    Frameworks & tools

    Specifically designed for Claude Code and Cursor-compatible agent architectures using Markdown-based agent definitions.

    Why this beats prompting it yourself

    Manually managing subagents often leads to "over-triggering" or agents with excessive permissions. This skill ensures a disciplined least-privilege approach and standardized trigger descriptions that keep your main context clean.

    Use cases

    • Create a dedicated security reviewer agent for PRs.
    • Build a log-parsing subagent to debug production traces.
    • Design a test-runner agent that autonomously fixes broken suites.
    • Split complex research tasks into smaller, parallelizable sub-tasks.

    Known limitations

    Does not support one-off tasks needing full conversation history. Requires manual file placement in .claude/agents/ directories.

    Designed for isolated tasks only. Not suitable for one-off steps or tasks requiring the full conversation context. User must manually save the generated Markdown to the agents directory.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 4 days ago

    • Free with an Agensi account

    Requires

    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed4 days ago
    Updated1 day ago

    What's inside

    Frequently Asked Questions

    This skill streamlines complex coding sessions by delegating repetitive tasks to specialized mini-agents, which prevents "context bloat" and keeps the primary AI model focused and responsive.

    It is specifically optimized for Claude Code and Cursor-compatible architectures that utilize Markdown-based (`.md`) agent definitions within local project directories.

    You receive a structured workflow that automates the creation of `.claude/agents/` files, including optimized system prompts, tool configurations, and trigger logic tailored for least-privilege access.

    Yes, once the skill generates the initial agent configuration, you can manually adjust the Markdown frontmatter or system instructions to fine-tune specific tool permissions or model selections.

    Subagents are best for isolated, repeatable tasks like code reviews or log analysis; they are not suitable for one-off tasks that require the full history of your primary conversation.

    New in Agents & Orchestration