Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create custom subagents for specialized AI tasks. Use when you want to create a new type of subagent, set up task-specific agents, configure code reviewers, debuggers, or domain-specific assistants with custom prompts.
.claude/skills/kunanonj-cursor-create-subagent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -30% | 0% |
This skill guides you through creating custom subagents for Cursor. Subagents are specialized AI assistants that run in isolated contexts with custom system prompts.
Subagents help you:
| Location | Scope | Priority | |----------|-------|----------| | .cursor/agents/ | Current project | Higher | | ~/.cursor/agents/ | All your projects | Lower |
When multiple subagents share the same name, the higher-priority location wins.
Create a .md file with YAML frontmatter and a markdown body (the system prompt):
markdown--- name: code-reviewer description: Reviews code for quality and best practices --- You are a code reviewer. When invoked, analyze the code and provide specific, actionable feedback on quality, security, and best practices.
| Field | Description | |-------|-------------| | name | Unique identifier (lowercase letters and hyphens only) | | description | When to delegate to this subagent (be specific!) |
yaml# Too vague description: Helps with code # Specific and actionable description: Expert code review specialist. Proactively reviews code for quality, security, and maintainability. Use immediately after writing or modifying code.
Include "use proactively" to encourage automatic delegation.
markdown--- name: code-reviewer description: Expert code review specialist. Proactively reviews code for quality, security, and maintainability. Use immediately after writing or modifying code. --- You are a senior code reviewer ensuring high standards of code quality and security. When invoked: 1. Run git diff to see recent changes 2. Focus on modified files 3. Begin review immediately Provide feedback organized by priority: - Critical issues (must fix) - Warnings (should fix) - Suggestions (consider improving)
markdown--- name: debugger description: Debugging specialist for errors, test failures, and unexpected behavior. Use proactively when encountering any issues. --- You are an expert debugger specializing in root cause analysis. When invoked: 1. Capture error message and stack trace 2. Identify reproduction steps 3. Isolate the failure location 4. Implement minimal fix 5. Verify solution works
markdown--- name: data-scientist description: Data analysis expert for SQL queries, BigQuery operations, and data insights. Use proactively for data analysis tasks and queries. --- You are a data scientist specializing in SQL and BigQuery analysis. When invoked: 1. Understand the data analysis requirement 2. Write efficient SQL queries 3. Analyze and summarize results 4. Present findings clearly
.cursor/agents/) or User-level (~/.cursor/agents/).md file with YAML frontmattername and description in frontmatter| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 14,233 | 7,738 | -46% | 1 | 1 | 0% | 2,656 | 2,221 | -16% | 0 | 0 | — |
case-01 | fail→pass | 12,959 | 6,421 | -50% | 1 | 1 | 0% | 2,310 | 2,039 | -12% | 0 | 0 | — |
case-02 | fail→pass | 15,512 | 7,613 | -51% | 1 | 1 | 0% | 2,912 | 2,290 | -21% | 0 | 0 | — |
case-04 | pass→pass | 8,893 | 6,684 | -25% | 1 | 1 | 0% | 1,701 | 2,147 | +26% | 0 | 0 | — |
case-05 | fail→fail | 5,053 | 4,458 | -12% | 1 | 1 | 0% | 1,031 | 1,704 | +65% | 0 | 0 | — |
case-06 | pass→pass | 9,589 | 8,430 | -12% | 1 | 1 | 0% | 1,918 | 2,246 | +17% | 0 | 0 | — |
case-07 | fail→pass | 10,407 | 6,763 | -35% | 1 | 1 | 0% | 1,919 | 2,081 | +8% | 0 | 0 | — |
case-08 | fail→pass | 12,508 | 4,376 | -65% | 1 | 1 | 0% | 2,460 | 1,721 | -30% | 0 | 0 | — |
case-09 | fail→pass | 10,117 | 3,848 | -62% | 1 | 1 | 0% | 1,968 | 1,599 | -19% | 0 | 0 | — |
case-10 | fail→pass | 8,153 | 2,504 | -69% | 1 | 1 | 0% | 1,508 | 1,286 | -15% | 0 | 0 | — |
case-11 | pass→pass | 3,574 | 2,139 | -40% | 1 | 1 | 0% | 705 | 1,242 | +76% | 0 | 0 | — |
case-12 | pass→pass | 11,566 | 6,011 | -48% | 1 | 1 | 0% | 1,679 | 1,904 | +13% | 0 | 0 | — |
case-13 | pass→pass | 8,087 | 3,091 | -62% | 1 | 1 | 0% | 1,529 | 1,445 | -5% | 0 | 0 | — |
case-14 | fail→pass | 10,500 | 4,965 | -53% | 1 | 1 | 0% | 1,893 | 1,784 | -6% | 0 | 0 | — |
case-15 | fail→pass | 9,119 | 5,042 | -45% | 1 | 1 | 0% | 1,766 | 1,837 | +4% | 0 | 0 | — |
case-16 | pass→pass | 5,571 | 2,418 | -57% | 1 | 1 | 0% | 1,017 | 1,296 | +27% | 0 | 0 | — |
case-17 | pass→pass | 3,629 | 3,363 | -7% | 1 | 1 | 0% | 643 | 1,505 | +134% | 0 | 0 | — |
case-18 | pass→pass | 10,480 | 4,684 | -55% | 1 | 1 | 0% | 1,931 | 1,810 | -6% | 0 | 0 | — |
case-19 | pass→pass | 9,378 | 4,105 | -56% | 1 | 1 | 0% | 1,722 | 1,613 | -6% | 0 | 0 | — |
case-20 | pass→pass | 10,806 | 7,992 | -26% | 1 | 1 | 0% | 1,925 | 2,351 | +22% | 0 | 0 | — |
case-21 | pass→pass | 14,578 | 4,134 | -72% | 1 | 1 | 0% | 2,657 | 1,612 | -39% | 0 | 0 | — |
case-22 | pass→pass | 8,698 | 2,673 | -69% | 1 | 1 | 0% | 1,237 | 1,330 | +8% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +41 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.