Custom Agent Usage

Custom Agent Usage

• Leverage native parallel subagent dispatch and 200k+ context windows where available.

Activation Conditions

Use symptom -> action triggers: when one matches, apply this skill and verify with the protocol below.

Activate this skill when:

• Discovering available custom agents from the local Claude or VS Code Insiders agent directories
• Understanding .agent.md file structure and frontmatter
• Checking if an agent can be invoked as a subagent
• Learning which agentName to use for delegation
• Understanding agent tools and capabilities

Custom Agent Discovery

Custom agents for this environment are discovered from these directories:

• Claude agents: C:\Users\LOQ\.claude\agents
• VS Code Insiders prompts: C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts

Notes:

• Filter to *.agent.md when discovering subagents.
• The VS Code Insiders prompts directory can also contain .prompt.md and .instructions.md files that are not subagents.

To discover agents: 1. Search for *.agent.md files in the configured Claude and VS Code Insiders directories. 2. Read frontmatter of each agent file to understand capabilities. 3. Ignore .prompt.md and .instructions.md files when the goal is subagent delegation.

Agent Frontmatter

Key frontmatter fields in .agent.md:

Invocability Check

CRITICAL: Only agents with disable-model-invocation: false in frontmatter can be invoked as subagents.

Check frontmatter:

---
name: "Code Explainer"
description: For analyzing and documenting existing code
disable-model-invocation: false  # Must be false for subagent delegation
tools: [Read, Search]
---

Using Custom Agents

Step 1: Discover Available Agents

Search the real agent directories and keep only *.agent.md files:

Get-ChildItem `
  'C:\Users\LOQ\.claude\agents', `
  'C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts' `
  -Filter *.agent.md `
  -File

Or use the helper script:

node .\scripts\agent-finder.js

Step 2: Check Invocability

Verify disable-model-invocation: false is set.

Step 3: Get Agent Name

Use the name field from frontmatter exactly as is.

• If frontmatter has name field value, use that value in quotes.
• If name is not specified, use the filename without the .agent.md extension.

Step 4: Delegate Task

runSubagent({
  agentName: "Playwright Tester Mode",  // Must match 'name' from frontmatter exactly
  description: "Test checkout flow",
  prompt: "Perform exploratory testing on the checkout flow: product selection -> cart -> payment confirmation. Generate comprehensive Playwright tests covering success scenarios, validation errors, edge cases (empty cart, payment failures), and accessibility."
})

Anti-Patterns

• Delegating or evaluating without a scoped success condition: The output becomes hard to review and easy to overbuild.
• Skipping the evidence step: A workflow that cannot be re-checked quickly is not ready for handoff.
• Bundling unrelated subtasks together: It creates noisy prompts, weaker ownership, and avoidable integration risk.

Verification Protocol

Before claiming "skill applied successfully":

1. Pass/fail: The Custom Agent Usage workflow names the agent boundary, delegated scope, and expected return artifact. 2. Pass/fail: Context passed to helpers is minimal, task-local, and free of hidden expected answers. 3. Pass/fail: Results are integrated only after evidence, diffs, or citations are checked by the controller. 4. Pressure-test scenario: Run the workflow on two similar tasks that must not share assumptions or leaked context. 5. Success metric: Zero context leakage; every delegated output is independently reviewable.

Workflow Example

// Step 1: Main agent analyzes task and identifies need for testing
// "I need comprehensive testing for the checkout flow"

// Step 2: Discover custom testing agent
// Found: C:\Users\LOQ\.claude\agents\playwright-tester.agent.md
// disable-model-invocation: false and name: "Playwright Tester Mode"

// Step 3: Delegate to custom agent
runSubagent({
  agentName: "Playwright Tester Mode",
  description: "Test checkout flow",
  prompt: "Perform exploratory testing on the checkout flow: product selection -> cart -> payment confirmation. Generate comprehensive Playwright tests covering success scenarios, validation errors, edge cases (empty cart, payment failures), and accessibility."
})

// Step 4: Review test output and integrate into test suite

Examples & Scripts

• Agent Discovery Workflow - Examples of finding and using custom agents
• Agent Finder Script - Node.js script to discover and inspect custom agents

<!-- PORTABILITY:START -->

Cross-Client Portability

This skill is written to stay usable across GitHub Copilot, Claude Code, Codex, and Gemini CLI.

• GitHub Copilot: keep the folder in a Copilot-visible skill or plugin path, or wrap the workflow as project instructions if the host does not support portable skill folders directly.
• Claude Code: keep the folder in a local skills directory or a compatible plugin or marketplace source.
• Codex: install or sync the folder into $CODEX_HOME/skills/<skill-name> and restart Codex after major changes.
• Gemini CLI: this repository generates a project command named /skills:custom-agent-usage from this skill. Rebuild commands with python scripts/export-gemini-skill.py custom-agent-usage and then run /commands reload inside Gemini CLI.

<!-- PORTABILITY:END -->

<!-- MCP:START -->

MCP Availability And Fallback

Preferred MCP Server: None required

• Fallback prompt: "Use the Custom Agent Usage skill without MCP. Rely on the local SKILL.md, bundled references or scripts, and manual verification. Show the exact commands, evidence, and final checks you used before concluding."
• If the current host does not expose a matching server, use the bundled references, scripts, native toolchain, and manual workflow already described in this skill.
• Treat direct local verification, rendered output, logs, tests, or screenshots as the fallback evidence path before completion.

<!-- MCP:END -->

Related Skills

• agent-task-mapping: Use it when the workflow also needs task-to-agent routing decisions.
• subagent-delegation: Use it when the workflow also needs safe, scoped delegation to helper agents.
• subagent-driven-development: Use it when the workflow also needs plan-driven implementation with reviewer loops.
• agentic-eval: Use it when the workflow also needs rubric-driven evaluation loops.

Changelog

[2026-04-25] - Version 1.2 Verification Protocol Refresh

Added

• Added a Verification Protocol section with skill-specific pass/fail checks, one pressure-test scenario, and a measurable success metric.
• Added guidance to leverage native parallel subagent dispatch and 200k+ context windows where available.

Changed

• Updated SKILL.md frontmatter to version: "1.2" and last_updated: 2026-04-25.
• Reframed activation guidance toward symptom -> action triggers and standardized two-stage review wording where applicable.

[2026-04-24] - Version 1.1 Refresh

Changed

• Updated the SKILL frontmatter version to 1.1 for the 2026-04-24 catalog refresh.

[2026-04-24] - Skill Refresh

Changed

• Standardized the SKILL frontmatter with version metadata, last-updated date, tags, and a concise catalog description.
• Reformatted the portability and MCP guidance with a preferred server line, a copy-paste fallback prompt, and consistent bullet lists.
• Added a catalog-standard Anti-Patterns section and refreshed the Related Skills links at the end of the skill.

[2026-04-24] - Catalog Audit Cleanup

Fixed

• Removed obsolete standalone Skill Paths guidance that duplicated the generated portability section.

All notable changes to this skill will be documented in this file.

[2026-04-04] - Cross-Client Portability Refresh

Changed

• Added a standard portability note covering GitHub Copilot, Claude Code, Codex, and Gemini CLI.
• Clarified that the core workflow does not require a dedicated MCP server and can run with local tools alone.

Tested

• Validated SKILL.md frontmatter, portability sections, and Gemini export readiness with python scripts/validate-skills.py.

[2026-03-09] - Custom Agent Discovery Correction

Changed

• Repointed custom-agent discovery guidance to the real local agent directories: C:\Users\LOQ\.claude\agents and C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts
• Updated the examples to search those directories directly instead of treating repo-local Copilot paths as the primary discovery roots

Fixed

• Removed the stale .copilot/agents and .github/copilot/agents discovery guidance
• Clarified that the VS Code Insiders prompts directory also contains .prompt.md and .instructions.md, so discovery must filter to *.agent.md
• Removed the external glob dependency from the helper script so it runs in the plain local Node environment used in this workspace

Tested

• Checked the real local directories on this machine and confirmed both exist
• Verified the correction against the files currently present in the Claude and VS Code Insiders agent directories

[2026-03-09] - Workspace Modernization

Changed

• Updated the workspace and global skill path guidance to match the current C:/Users/LOQ/.agents/skills/ fallback path
• Removed duplicate related-skill content so the skill reads cleanly

[2026-02-28] - Description Rewrite & Cross-References

Changed

• Rewrote skill description to ~200 characters with clear, specific activation keywords
• Improved keyword specificity to reduce overlap with related skills

Added

• ## Related Skills cross-reference table with 2-4 related skills and "Use When" guidance

Custom Agent Usage Examples

Example 1: Finding and Using a Testing Agent

Step 1: Discover Available Agents

Search the real agent directories and keep only *.agent.md files:

Get-ChildItem `
  'C:\Users\LOQ\.claude\agents', `
  'C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts' `
  -Filter *.agent.md `
  -File

Step 2: Check Agent Frontmatter

Read the .agent.md file to verify it can be invoked:

---
name: "Playwright Tester Mode"
description: Comprehensive testing with exploratory testing, Playwright test generation
disable-model-invocation: false  # Must be false for subagent delegation
tools: [Read, Write, Bash, WebSearch]
---

Step 3: Use the Agent

// Delegate with exact name from frontmatter
runSubagent({
  agentName: "Playwright Tester Mode",
  description: "Test checkout flow",
  prompt: "Perform exploratory testing on the checkout flow: product selection -> cart -> payment confirmation. Generate comprehensive Playwright tests covering success scenarios, validation errors, edge cases (empty cart, payment failures), and accessibility."
});

Example 2: Finding and Using a Documentation Agent

Discovery Workflow

// 1. Find all .agent.md files in the configured directories
const agentDirs = [
  'C:\\Users\\LOQ\\.claude\\agents',
  'C:\\Users\\LOQ\\AppData\\Roaming\\Code - Insiders\\User\\prompts'
];
const agentFiles = agentDirs.flatMap(findAgentFilesInDirectory);

function findAgentFilesInDirectory(rootDir) {
  return fs.readdirSync(rootDir)
    .filter(name => name.endsWith('.agent.md'))
    .map(name => path.join(rootDir, name));
}

// 2. Read each file's frontmatter
for (const file of agentFiles) {
  const content = readFileSync(file, 'utf8');
  const frontmatter = parseFrontmatter(content);

  console.log({
    name: frontmatter.name,
    description: frontmatter.description,
    invocable: frontmatter.disableModelInvocation === false
  });
}

// 3. Select appropriate agent
// Found: Tech Writer with disable-model-invocation: false

// 4. Delegate task
runSubagent({
  agentName: "Tech Writer",
  description: "Create API documentation",
  prompt: "Create comprehensive API documentation for the authentication endpoints. Include request/response schemas, authentication requirements, error codes, and usage examples."
});

Example 3: Checking Invocability

/**
 * Check if an agent can be invoked as a subagent
 */
function isAgentInvocable(agentFilePath) {
  const content = fs.readFileSync(agentFilePath, 'utf8');
  const match = content.match(/disable-model-invocation:\s*(true|false)/);

  if (!match) {
    return false; // Cannot determine if invocable
  }

  return match[1] === 'false';
}

// Usage
const playwrightTester = 'C:\\Users\\LOQ\\.claude\\agents\\playwright-tester.agent.md';
if (isAgentInvocable(playwrightTester)) {
  console.log('This agent can be invoked as a subagent');
} else {
  console.log('This agent cannot be invoked as a subagent');
}

Example 4: Getting Agent Name

/**
 * Extract agent name from .agent.md frontmatter
 */
function getAgentName(agentFilePath) {
  const content = fs.readFileSync(agentFilePath, 'utf8');

  // Try to extract name from frontmatter
  const nameMatch = content.match(/name:\s*["']([^"']+)["']/);
  if (nameMatch) {
    return nameMatch[1]; // Use name from frontmatter
  }

  // Fallback to filename
  return path.basename(agentFilePath)
    .replace('.agent.md', '')
    .replace('.md', '');
}

// Usage
const agentName = getAgentName('C:\\Users\\LOQ\\.claude\\agents\\Code-Explainer.agent.md');
// Returns: "Code Explainer" (from frontmatter)
// or: "Code-Explainer" (from filename if no frontmatter name)

Example 5: Complete Workflow

/**
 * Find and use appropriate agent for a task
 */
function delegateToAppropriateAgent(taskDescription) {
  // 1. Find all agent files
  const agentDirs = [
    'C:\\Users\\LOQ\\.claude\\agents',
    'C:\\Users\\LOQ\\AppData\\Roaming\\Code - Insiders\\User\\prompts'
  ];
  const agentFiles = agentDirs.flatMap(findAgentFilesInDirectory);

  // 2. Extract agent info
  const agents = agentFiles.map(file => ({
    file,
    name: getAgentName(file),
    description: extractDescription(file),
    invocable: isAgentInvocable(file)
  }));

  // 3. Filter invocable agents
  const invocableAgents = agents.filter(a => a.invocable);

  // 4. Find matching agent (simplified matching logic)
  const matchingAgent = invocableAgents.find(agent =>
    taskDescription.toLowerCase().includes(agent.description.toLowerCase().split(' ')[0])
  );

  if (!matchingAgent) {
    console.log('No matching agent found');
    return;
  }

  // 5. Delegate to agent
  runSubagent({
    agentName: matchingAgent.name,
    description: taskDescription,
    prompt: `Handle this task: ${taskDescription}`
  });
}

// Usage
delegateToAppropriateAgent("Test the shopping cart functionality");
// Would find and delegate to "Playwright Tester Mode"

Custom Agent Discovery Workflow

Frontmatter Structure

Proper .agent.md frontmatter:

---
name: "Display Name"
description: Brief description of what this agent does
disable-model-invocation: false  # Required: must be false for subagent delegation
tools: [List of tools available]
---

Key Fields Explained

name

• Purpose: Display name used when calling runSubagent
• Required: Not strictly required, but highly recommended for clear delegation
• Usage: Use this exact value in agentName parameter
• Format: Can include spaces, should be descriptive

description

• Purpose: Explains what the agent specializes in
• Required: Recommended for agent discoverability
• Content: Should describe purpose and use cases

disable-model-invocation

• Purpose: Controls whether agent can be invoked via runSubagent
• Required: Yes, for subagent delegation
• Value: Must be false for subagent invocation
• When `true`: Agent cannot be delegated to directly

tools

• Purpose: Lists tools available to the agent
• Required: Optional
• Format: Array of tool names
• Usage: Helps understand agent capabilities

Discoverable Agents

Examples of properly configured agents:

---
name: "Code Explainer"
description: For analyzing and documenting existing code
disable-model-invocation: false
tools: [Read, Search, Symbol]
---

---
name: "Playwright Tester Mode"
description: For comprehensive web application testing
disable-model-invocation: false
tools: [Browser, Page, Locator]
---

Search Patterns

Command Line

# Claude Code agent directory
Get-ChildItem 'C:\Users\LOQ\.claude\agents' -Filter *.agent.md -File

# VS Code Insiders prompts directory
# Keep only *.agent.md because this folder can also contain .prompt.md and .instructions.md
Get-ChildItem 'C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts' -Filter *.agent.md -File

VS Code Search

• Use file search with pattern: *.agent.md
• Search inside:
• C:\Users\LOQ\.claude\agents
• C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts
• Exclude: node_modules/**

Validation Checklist

Before using an agent as a subagent, verify:

• [ ] File follows .agent.md naming convention
• [ ] Has valid YAML frontmatter with --- delimiters
• [ ] disable-model-invocation is explicitly set to false
• [ ] name field is present and meaningful
• [ ] description explains agent's purpose clearly
• [ ] File is in a discoverable directory: C:\Users\LOQ\.claude\agents or C:\Users\LOQ\AppData\Roaming\Code - Insiders\User\prompts