Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create and maintain Claude Code agents/subagents (.claude/agents/*.md) with YAML frontmatter (name/description/tools/model/permissionMode/skills/hooks), least-privilege tool selection, delegation patterns (Task), context budgeting, and safety best practices.
.claude/skills/majiayu000-claude-code-agents/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 66% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -47% | 0% |
Create and maintain Claude Code agents/subagents with predictable behavior, least-privilege tools, and explicit delegation contracts.
.claude/agents/<agent-name>.md (kebab-case filename).name, description; optional: tools, model, permissionMode, skills, hooks).Minimal template:
markdown--- name: sql-optimizer description: Optimize SQL queries, explain tradeoffs, and propose safe indexes tools: Read, Grep, Glob model: sonnet --- # SQL Optimizer ## Responsibilities - Diagnose bottlenecks using query shape and plans when available - Propose optimizations with risks and expected impact ## Workflow 1. Identify the slow path and data volume assumptions 2. Propose changes (query rewrite, indexes, stats) with rationale 3. Provide a verification plan ## Output Contract - Summary (1–3 bullets) - Recommendations (ordered) - Verification (commands/tests to run)
sonnet for most work).Edit/Write unless required.Task, define a handoff contract (inputs, constraints, output format).name (REQUIRED): kebab-case; match filename (without .md).description (REQUIRED): state when to invoke + what it does; include keywords users will say.tools (OPTIONAL): explicit allow-list; prefer small, purpose-built sets.model (OPTIONAL): haiku for fast checks, sonnet for most tasks, opus for high-stakes reasoning, inherit to match parent.permissionMode (OPTIONAL): prefer defaults; change only with a clear reason and understand the tradeoffs.skills (OPTIONAL): preload skill packs for domain expertise; keep the list minimal.hooks (OPTIONAL): automate guardrails; prefer using the hooks skill for patterns and safety.For full tool semantics and permission patterns, use references/agent-tools.md. For orchestration and anti-patterns, use references/agent-patterns.md.
CLAUDE.md/project memory and domain knowledge in skills.data/sources.json.name matches filename; description is single-line and trigger-oriented; tools are minimal; model fits risk.frameworks/shared-skills/skills/claude-code-agents/references/agent-patterns.mdframeworks/shared-skills/skills/claude-code-agents/references/agent-tools.mdframeworks/shared-skills/skills/claude-code-agents/data/sources.jsonframeworks/shared-skills/skills/claude-code-skills/SKILL.mdframeworks/shared-skills/skills/claude-code-commands/SKILL.mdframeworks/shared-skills/skills/claude-code-hooks/SKILL.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-12 | pass→pass | 17,848 | 12,077 | -32% | 1 | 1 | 0% | 2,579 | 2,922 | +13% | 0 | 0 | — |
case-01 | fail→pass | 15,712 | 18,091 | +15% | 1 | 1 | 0% | 2,435 | 4,030 | +66% | 0 | 0 | — |
case-02 | fail→pass | 16,792 | 15,677 | -7% | 1 | 1 | 0% | 2,642 | 3,490 | +32% | 0 | 0 | — |
case-03 | fail→pass | 16,942 | 15,814 | -7% | 1 | 1 | 0% | 2,726 | 3,629 | +33% | 0 | 0 | — |
case-04 | pass→pass | 9,799 | 5,463 | -44% | 1 | 1 | 0% | 1,567 | 1,813 | +16% | 0 | 0 | — |
case-05 | fail→pass | 13,554 | 4,697 | -65% | 1 | 1 | 0% | 2,162 | 1,764 | -18% | 0 | 0 | — |
case-06 | pass→pass | 30,635 | 4,628 | -85% | 1 | 1 | 0% | 1,930 | 1,658 | -14% | 0 | 0 | — |
case-07 | pass→pass | 7,709 | 3,470 | -55% | 1 | 1 | 0% | 1,248 | 1,491 | +19% | 0 | 0 | — |
case-08 | pass→pass | 7,972 | 19,748 | +148% | 1 | 1 | 0% | 1,219 | 1,324 | +9% | 0 | 0 | — |
case-09 | fail→pass | 18,353 | 3,758 | -80% | 1 | 1 | 0% | 2,867 | 1,524 | -47% | 0 | 0 | — |
case-10 | pass→pass | 8,041 | 2,161 | -73% | 1 | 1 | 0% | 1,186 | 1,293 | +9% | 0 | 0 | — |
case-11 | pass→pass | 12,615 | 5,279 | -58% | 1 | 1 | 0% | 1,925 | 1,764 | -8% | 0 | 0 | — |
case-13 | pass→pass | 11,910 | 3,112 | -74% | 1 | 1 | 0% | 1,731 | 1,389 | -20% | 0 | 0 | — |
case-14 | pass→pass | 14,888 | 10,462 | -30% | 1 | 1 | 0% | 2,200 | 2,516 | +14% | 0 | 0 | — |
case-15 | pass→pass | 5,148 | 2,092 | -59% | 1 | 1 | 0% | 848 | 1,334 | +57% | 0 | 0 | — |
case-16 | pass→pass | 15,364 | 6,534 | -57% | 1 | 1 | 0% | 2,162 | 1,934 | -11% | 0 | 0 | — |
case-17 | pass→pass | 15,809 | 9,107 | -42% | 1 | 1 | 0% | 2,293 | 2,376 | +4% | 0 | 0 | — |
case-18 | fail→pass | 14,707 | 8,215 | -44% | 1 | 1 | 0% | 2,105 | 2,206 | +5% | 0 | 0 | — |
case-19 | fail→pass | 18,871 | 11,655 | -38% | 1 | 1 | 0% | 3,109 | 2,704 | -13% | 0 | 0 | — |
case-20 | pass→pass | 14,588 | 12,321 | -16% | 1 | 1 | 0% | 2,362 | 2,887 | +22% | 0 | 0 | — |
case-21 | pass→fail | 18,003 | 21,941 | +22% | 1 | 1 | 0% | 3,289 | 5,468 | +66% | 0 | 0 | — |
case-22 | pass→pass | 19,917 | 16,296 | -18% | 1 | 1 | 0% | 3,409 | 3,298 | -3% | 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 +27 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.