# 📊 Agent Skill Benchmark Report

> Generated: 2026-04-11T12:00:22.838Z
> Token counting: `ceil(characters / 4)` — cl100k_base approximation.
> Baselines: derived from **real, measured example prompts** (see Methodology).
> Quality: structural rubric (0–10), no live LLM calls required.

## ❓ How to Read This Report

This benchmark answers: **"How many tokens and dollars does an agent skill save compared to a developer writing the same guidance inline?"**

**WITHOUT a skill**: A developer writes domain knowledge directly into the prompt every time (Baseline).
**WITH a skill**: The agent loads the SKILL.md file (~400 tokens) — structured, reusable, cached.

**Eval Alignment**: % of eval assertion values that appear in SKILL.md. High alignment means the skill actually teaches what the evals test — the static proxy for "with skill > without skill" behavioral improvement.

## 🔢 Executive Summary

| Metric                            | Value                             |
| --------------------------------- | --------------------------------- |
| Total Skills Benchmarked          | **237**           |
| Avg. Tokens WITH Skill (SKILL.md) | **516 tokens**    |
| Baseline: Light prompt (no skill) | **1449 tokens** ↓ see Methodology |
| Baseline: Heavy prompt (no skill) | **3656 tokens** ↓ see Methodology |
| Avg. Token Savings vs Light       | **64%** (933 tokens/call) |
| Avg. Token Savings vs Heavy       | **86%** (3140 tokens/call) |
| Avg. Quality Score                | **9.9/10** |
| Skills with Evals                 | **237 / 237** |
| Avg. Eval Alignment               | **94%** (eval assertions covered by SKILL.md) |

## 📜 History

| Version | Date       | Skills | Avg Tokens | Savings (%) | Quality | Report |
| ------- | ---------- | ------ | ---------- | ----------- | ------- | ------ |
| v2.1.1 | 2026-04-11 | 237 | 516 | 86% | 9.9/10 | [Full Report](benchmarks/archive/v2.1.1.md) |
| v2.1.0 | 2026-04-04 | 237 | 526 | 86% | 9.9/10 | [Full Report](benchmarks/archive/v2.1.0.md) |
| v2.0.1 | 2026-03-30 | 238 | 527 | 86% | 9.8/10 | [Full Report](benchmarks/archive/v2.0.1.md) |
| v2.0.0 | 2026-03-25 | 235 | 523 | 86% | 9.9/10 | [Full Report](benchmarks/archive/v2.0.0.md) |
| v1.10.3 | 2026-03-21 | 234 | 505 | 86% | 9.8/10 | [Full Report](benchmarks/archive/v1.10.3.md) |
| v1.10.1 | 2026-03-16 | 229 | 428 | 88% | 9.9/10 | [Full Report](benchmarks/archive/v1.10.1.md) |
| v1.10.0 | 2026-03-16 | 229 | 434 | 88% | 7/10 | [Full Report](benchmarks/archive/v1.10.0.md) |
| v1.9.3 | 2026-03-15 | 229 | 460 | 87% | 8.9/10 | [Full Report](benchmarks/archive/v1.9.3.md) |
| v1.9.2 | 2026-03-07 | 228 | 458 | 87% | 8.9/10 | [Full Report](benchmarks/archive/v1.9.2.md) |
| v1.9.1 | 2026-03-07 | 228 | 458 | 87% | 8.9/10 | [Full Report](benchmarks/archive/v1.9.1.md) |
| v1.9.0 | 2026-03-05 | 228 | 457 | 88% | 8.9/10 | [Full Report](benchmarks/archive/v1.9.0.md) |
| v1.8.0 | 2026-03-02 | 228 | 443 | 88% | 8.9/10 | [Full Report](benchmarks/archive/v1.8.0.md) |
| v1.7.3 | 2026-02-25 | 222 | 418 | 89% | 8.9/10 | [Full Report](benchmarks/archive/v1.7.3.md) |
| v1.7.2 | 2026-02-25 | 220 | 413 | 89% | 8.9/10 | [Full Report](benchmarks/archive/v1.7.2.md) |

### 💰 Cost Comparison — Per Single Call (Average Skill)

> Comparison based on **Heavy Baseline** vs. modern and speculative models.

| Model             | Original Cost | Skill Cost | Net Savings    | % Saved |
| ----------------- | ------------- | ---------- | -------------- | ------- |
| Gemini 3 Flash    | $0.0018280    | $0.0002580 | **$0.0015700    ** | 86% |
| GPT-5             | $0.0045700    | $0.0006450 | **$0.0039250    ** | 86% |
| Gemini 3.1 Pro    | $0.0073120    | $0.0010320 | **$0.0062800    ** | 86% |
| Claude Sonnet 4.5 | $0.0109680    | $0.0015480 | **$0.0094200    ** | 86% |

### 📈 Monthly Savings at Scale — (Avg Skill vs Heavy Prompt)

| Daily Calls | Original Cost/mo | Monthly Savings (1 skill) | Monthly Savings (50 skills) | Model |
| ----------- | ---------------- | ------------------------- | --------------------------- | ----- |
| 1,000       | $137.1000       /mo | $117.7500               /mo | $5887.5000                /mo | GPT-5 |
| 1,000       | $329.0400       /mo | $282.6000               /mo | $14130.0000               /mo | Claude Sonnet 4.5 |
| 1,000       | $219.3600       /mo | $188.4000               /mo | $9420.0000                /mo | Gemini 3.1 Pro |

## 📦 Per-Category Summary

<details>
<summary><h3>📦 android (22 skills | avg 344 tokens | quality 10.0/10 | eval alignment 93%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `android-architecture ` | 504    | █████████░ 86%     | 10/10 | 3 | ✅ 88% |
| `android-background-work` | 302    | █████████░ 92%     | 10/10 | 3 | ✅ 100% |
| `android-compose      ` | 447    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `android-concurrency  ` | 314    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |
| `android-deployment   ` | 328    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `android-design-system` | 252    | █████████░ 93%     | 10/10 | 3 | ✅ 100% |
| `android-di           ` | 309    | █████████░ 92%     | 10/10 | 3 | ✅ 88% |
| `android-legacy-navigation` | 310    | █████████░ 92%     | 10/10 | 3 | ✅ 86% |
| `android-legacy-security` | 444    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `android-legacy-state ` | 256    | █████████░ 93%     | 10/10 | 3 | ✅ 100% |
| `android-navigation   ` | 261    | █████████░ 93%     | 10/10 | 3 | ✅ 100% |
| `android-navigation-type-safe` | 267    | █████████░ 93%     | 10/10 | 3 | ✅ 83% |
| `android-networking   ` | 409    | █████████░ 89%     | 10/10 | 3 | ✅ 83% |
| `android-notifications` | 415    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `android-performance  ` | 384    | █████████░ 89%     | 10/10 | 3 | ✅ 88% |
| `android-persistence  ` | 297    | █████████░ 92%     | 10/10 | 3 | ✅ 100% |
| `android-resources    ` | 409    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `android-security     ` | 397    | █████████░ 89%     | 10/10 | 3 | ✅ 86% |
| `android-state        ` | 361    | █████████░ 90%     | 10/10 | 3 | ✅ 88% |
| `android-testing      ` | 318    | █████████░ 91%     | 10/10 | 3 | ✅ 88% |
| `android-tooling      ` | 294    | █████████░ 92%     | 10/10 | 3 | ✅ 88% |
| `android-xml-views    ` | 295    | █████████░ 92%     | 10/10 | 3 | ✅ 88% |

</details>

<details>
<summary><h3>📦 angular (15 skills | avg 497 tokens | quality 9.9/10 | eval alignment 87%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `angular-architecture ` | 613    | ████████░░ 83%     | 10/10 | 6 | ✅ 95% |
| `angular-components   ` | 613    | ████████░░ 83%     | 10/10 | 9 | ✅ 77% |
| `angular-dependency-injection` | 518    | █████████░ 86%     | 10/10 | 6 | ✅ 89% |
| `angular-directives-pipes` | 489    | █████████░ 87%     | 10/10 | 6 | ✅ 91% |
| `angular-forms        ` | 344    | █████████░ 91%     | 10/10 | 6 | ✅ 100% |
| `angular-http-client  ` | 557    | █████████░ 85%     | 10/10 | 6 | ✅ 92% |
| `angular-performance  ` | 472    | █████████░ 87%     | 10/10 | 6 | ✅ 82% |
| `angular-routing      ` | 380    | █████████░ 90%     | 10/10 | 6 | ✅ 100% |
| `angular-rxjs-interop ` | 501    | █████████░ 86%     | 10/10 | 6 | ✅ 95% |
| `angular-security     ` | 494    | █████████░ 86%     | 10/10 | 6 | ✅ 78% |
| `angular-ssr          ` | 472    | █████████░ 87%     | 10/10 | 6 | ✅ 90% |
| `angular-state-management` | 402    | █████████░ 89%     | 10/10 | 6 | ✅ 81% |
| `angular-style-guide  ` | 500    | █████████░ 86%     | 10/10 | 6 | ✅ 71% |
| `angular-testing      ` | 422    | █████████░ 88%     | 10/10 | 6 | ✅ 70% |
| `angular-tooling      ` | 672    | ████████░░ 82%     | 8/10 | 6 | ✅ 100% |

</details>

<details>
<summary><h3>📦 common (31 skills | avg 613 tokens | quality 9.9/10 | eval alignment 94%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `common-architecture-audit` | 620    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `common-architecture-diagramming` | 452    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `common-best-practices` | 384    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `common-code-review   ` | 346    | █████████░ 91%     | 10/10 | 3 | ✅ 83% |
| `common-context-optimization` | 555    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `common-dast-tooling  ` | 579    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `common-debugging     ` | 352    | █████████░ 90%     | 10/10 | 3 | ✅ 83% |
| `common-documentation ` | 349    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `common-error-handling` | 392    | █████████░ 89%     | 10/10 | 3 | ✅ 83% |
| `common-feedback-reporter` | 859    | ████████░░ 77%     | 10/10 | 4 | ✅ 100% |
| `common-git-collaboration` | 505    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `common-learning-log  ` | 574    | ████████░░ 84%     | 10/10 | 3 | ✅ 83% |
| `common-llm-security  ` | 681    | ████████░░ 81%     | 10/10 | 3 | ✅ 100% |
| `common-mobile-animation` | 522    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `common-mobile-ux-core` | 372    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `common-observability ` | 378    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `common-owasp         ` | 894    | ████████░░ 76%     | 10/10 | 3 | ✅ 100% |
| `common-performance-engineering` | 674    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `common-product-requirements` | 422    | █████████░ 88%     | 10/10 | 3 | ✅ 83% |
| `common-protocol-enforcement` | 398    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `common-security-audit` | 856    | ████████░░ 77%     | 10/10 | 3 | ✅ 83% |
| `common-security-standards` | 705    | ████████░░ 81%     | 10/10 | 3 | ✅ 83% |
| `common-session-retrospective` | 655    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `common-skill-creator ` | 1336   | ██████░░░░ 63%     | 10/10 | 3 | ✅ 100% |
| `common-store-changelog` | 684    | ████████░░ 81%     | 10/10 | 4 | ⚠️ 43% |
| `common-system-design ` | 711    | ████████░░ 81%     | 10/10 | 3 | ✅ 100% |
| `common-tdd           ` | 634    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `common-ui-design     ` | 774    | ████████░░ 79%     | 10/10 | 3 | ✅ 100% |
| `common-workflow-writing` | 550    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `common-accessibility ` | 994    | ███████░░░ 73%     | 8/10 | 3 | ✅ 83% |
| `common-api-design    ` | 805    | ████████░░ 78%     | 8/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 dart (3 skills | avg 530 tokens | quality 10.0/10 | eval alignment 95%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `dart-best-practices  ` | 487    | █████████░ 87%     | 10/10 | 3 | ✅ 100% |
| `dart-language        ` | 612    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `dart-tooling         ` | 490    | █████████░ 87%     | 10/10 | 3 | ✅ 86% |

</details>

<details>
<summary><h3>📦 database (3 skills | avg 555 tokens | quality 10.0/10 | eval alignment 95%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `database-mongodb     ` | 617    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `database-postgresql  ` | 451    | █████████░ 88%     | 10/10 | 3 | ✅ 86% |
| `database-redis       ` | 596    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 flutter (21 skills | avg 497 tokens | quality 9.7/10 | eval alignment 93%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `flutter-cicd         ` | 512    | █████████░ 86%     | 10/10 | 3 | ✅ 89% |
| `flutter-design-system` | 504    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `flutter-error-handling` | 576    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `flutter-feature-based-clean-architecture` | 576    | ████████░░ 84%     | 10/10 | 3 | ✅ 88% |
| `flutter-getx-navigation` | 325    | █████████░ 91%     | 10/10 | 3 | ✅ 86% |
| `flutter-getx-state-management` | 435    | █████████░ 88%     | 10/10 | 3 | ✅ 86% |
| `flutter-go-router-navigation` | 546    | █████████░ 85%     | 10/10 | 3 | ✅ 89% |
| `flutter-idiomatic-flutter` | 337    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |
| `flutter-layer-based-clean-architecture` | 643    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `flutter-performance  ` | 428    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `flutter-retrofit-networking` | 543    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `flutter-riverpod-state-management` | 532    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `flutter-testing      ` | 746    | ████████░░ 80%     | 10/10 | 3 | ✅ 100% |
| `flutter-widgets      ` | 446    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `flutter-auto-route-navigation` | 469    | █████████░ 87%     | 9/10 | 3 | ✅ 100% |
| `flutter-bloc-state-management` | 602    | ████████░░ 84%     | 9/10 | 3 | ✅ 100% |
| `flutter-dependency-injection` | 497    | █████████░ 86%     | 9/10 | 3 | ✅ 80% |
| `flutter-localization ` | 486    | █████████░ 87%     | 9/10 | 3 | ✅ 100% |
| `flutter-navigation   ` | 386    | █████████░ 89%     | 9/10 | 3 | ✅ 100% |
| `flutter-notifications` | 389    | █████████░ 89%     | 9/10 | 3 | ✅ 100% |
| `flutter-security     ` | 459    | █████████░ 87%     | 9/10 | 3 | ⚠️ 50% |

</details>

<details>
<summary><h3>📦 golang (11 skills | avg 447 tokens | quality 10.0/10 | eval alignment 95%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `golang-api-server    ` | 443    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `golang-architecture  ` | 503    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `golang-concurrency   ` | 424    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `golang-configuration ` | 430    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `golang-database      ` | 449    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `golang-error-handling` | 341    | █████████░ 91%     | 10/10 | 3 | ✅ 83% |
| `golang-language      ` | 496    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `golang-logging       ` | 389    | █████████░ 89%     | 10/10 | 3 | ✅ 83% |
| `golang-security      ` | 509    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `golang-testing       ` | 417    | █████████░ 89%     | 10/10 | 3 | ✅ 83% |
| `golang-tooling       ` | 514    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 ios (15 skills | avg 363 tokens | quality 10.0/10 | eval alignment 96%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `ios-app-lifecycle    ` | 339    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |
| `ios-architecture     ` | 645    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `ios-dependency-injection` | 311    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |
| `ios-deployment       ` | 339    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |
| `ios-design-system    ` | 239    | █████████░ 93%     | 10/10 | 3 | ✅ 100% |
| `ios-localization     ` | 370    | █████████░ 90%     | 10/10 | 3 | ✅ 89% |
| `ios-navigation       ` | 281    | █████████░ 92%     | 10/10 | 3 | ✅ 100% |
| `ios-networking       ` | 369    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `ios-notifications    ` | 292    | █████████░ 92%     | 10/10 | 3 | ✅ 100% |
| `ios-performance      ` | 362    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `ios-persistence      ` | 345    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `ios-security         ` | 387    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `ios-state-management ` | 350    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `ios-swiftui          ` | 401    | █████████░ 89%     | 10/10 | 3 | ✅ 88% |
| `ios-ui-navigation    ` | 416    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 java (5 skills | avg 474 tokens | quality 10.0/10 | eval alignment 98%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `java-best-practices  ` | 461    | █████████░ 87%     | 10/10 | 3 | ✅ 89% |
| `java-concurrency     ` | 427    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `java-language        ` | 515    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `java-testing         ` | 516    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `java-tooling         ` | 450    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 javascript (3 skills | avg 348 tokens | quality 10.0/10 | eval alignment 93%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `javascript-best-practices` | 283    | █████████░ 92%     | 10/10 | 3 | ✅ 89% |
| `javascript-language  ` | 432    | █████████░ 88%     | 10/10 | 3 | ✅ 90% |
| `javascript-tooling   ` | 330    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 kotlin (4 skills | avg 397 tokens | quality 10.0/10 | eval alignment 95%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `kotlin-best-practices` | 448    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `kotlin-coroutines    ` | 380    | █████████░ 90%     | 10/10 | 3 | ✅ 89% |
| `kotlin-language      ` | 427    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `kotlin-tooling       ` | 333    | █████████░ 91%     | 10/10 | 3 | ✅ 89% |

</details>

<details>
<summary><h3>📦 laravel (10 skills | avg 646 tokens | quality 10.0/10 | eval alignment 90%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `laravel-api          ` | 702    | ████████░░ 81%     | 10/10 | 6 | ✅ 88% |
| `laravel-architecture ` | 391    | █████████░ 89%     | 10/10 | 6 | ✅ 100% |
| `laravel-background-processing` | 617    | ████████░░ 83%     | 10/10 | 6 | ✅ 91% |
| `laravel-clean-architecture` | 658    | ████████░░ 82%     | 10/10 | 6 | ✅ 76% |
| `laravel-database-expert` | 699    | ████████░░ 81%     | 10/10 | 6 | ✅ 84% |
| `laravel-eloquent     ` | 621    | ████████░░ 83%     | 10/10 | 6 | ✅ 85% |
| `laravel-security     ` | 724    | ████████░░ 80%     | 10/10 | 6 | ✅ 95% |
| `laravel-sessions-middleware` | 674    | ████████░░ 82%     | 10/10 | 6 | ✅ 90% |
| `laravel-testing      ` | 706    | ████████░░ 81%     | 10/10 | 6 | ✅ 95% |
| `laravel-tooling      ` | 671    | ████████░░ 82%     | 10/10 | 6 | ✅ 100% |

</details>

<details>
<summary><h3>📦 nestjs (21 skills | avg 610 tokens | quality 9.9/10 | eval alignment 98%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `nestjs-api-standards ` | 591    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `nestjs-architecture  ` | 542    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `nestjs-bullmq        ` | 896    | ████████░░ 75%     | 10/10 | 3 | ✅ 100% |
| `nestjs-caching       ` | 586    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `nestjs-configuration ` | 579    | ████████░░ 84%     | 10/10 | 3 | ✅ 83% |
| `nestjs-database      ` | 649    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `nestjs-deployment    ` | 678    | ████████░░ 81%     | 10/10 | 3 | ✅ 100% |
| `nestjs-documentation ` | 526    | █████████░ 86%     | 10/10 | 3 | ✅ 83% |
| `nestjs-error-handling` | 565    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `nestjs-file-uploads  ` | 396    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `nestjs-notification  ` | 502    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `nestjs-observability ` | 440    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `nestjs-performance   ` | 947    | ███████░░░ 74%     | 10/10 | 3 | ✅ 100% |
| `nestjs-real-time     ` | 877    | ████████░░ 76%     | 10/10 | 3 | ✅ 100% |
| `nestjs-scheduling    ` | 549    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `nestjs-search        ` | 500    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `nestjs-security      ` | 758    | ████████░░ 79%     | 10/10 | 3 | ✅ 100% |
| `nestjs-security-isolation` | 526    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `nestjs-testing       ` | 555    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `nestjs-transport     ` | 429    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `nestjs-controllers-services` | 715    | ████████░░ 80%     | 8/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 nextjs (18 skills | avg 638 tokens | quality 9.9/10 | eval alignment 97%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `nextjs-app-router    ` | 974    | ███████░░░ 73%     | 10/10 | 3 | ✅ 100% |
| `nextjs-architecture  ` | 1021   | ███████░░░ 72%     | 10/10 | 3 | ✅ 100% |
| `nextjs-authentication` | 488    | █████████░ 87%     | 10/10 | 3 | ✅ 100% |
| `nextjs-caching       ` | 783    | ████████░░ 79%     | 10/10 | 3 | ✅ 100% |
| `nextjs-data-access-layer` | 519    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `nextjs-data-fetching ` | 448    | █████████░ 88%     | 10/10 | 6 | ✅ 100% |
| `nextjs-i18n          ` | 585    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `nextjs-optimization  ` | 511    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `nextjs-rendering     ` | 730    | ████████░░ 80%     | 10/10 | 3 | ✅ 100% |
| `nextjs-security      ` | 670    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `nextjs-server-actions` | 718    | ████████░░ 80%     | 10/10 | 6 | ✅ 85% |
| `nextjs-server-components` | 625    | ████████░░ 83%     | 10/10 | 6 | ✅ 78% |
| `nextjs-state-management` | 584    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `nextjs-styling       ` | 622    | ████████░░ 83%     | 10/10 | 3 | ✅ 83% |
| `nextjs-testing       ` | 627    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `nextjs-tooling       ` | 390    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `nextjs-upgrade       ` | 551    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `nextjs-pages-router  ` | 646    | ████████░░ 82%     | 9/10 | 6 | ✅ 100% |

</details>

<details>
<summary><h3>📦 php (7 skills | avg 509 tokens | quality 9.6/10 | eval alignment 93%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `php-best-practices   ` | 518    | █████████░ 86%     | 10/10 | 6 | ✅ 92% |
| `php-security         ` | 537    | █████████░ 85%     | 10/10 | 6 | ✅ 100% |
| `php-testing          ` | 531    | █████████░ 85%     | 10/10 | 6 | ✅ 83% |
| `php-tooling          ` | 530    | █████████░ 86%     | 10/10 | 6 | ✅ 86% |
| `php-concurrency      ` | 520    | █████████░ 86%     | 9/10 | 6 | ✅ 100% |
| `php-error-handling   ` | 471    | █████████░ 87%     | 9/10 | 6 | ✅ 100% |
| `php-language         ` | 454    | █████████░ 88%     | 9/10 | 6 | ✅ 91% |

</details>

<details>
<summary><h3>📦 quality-engineering (5 skills | avg 754 tokens | quality 10.0/10 | eval alignment 95%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `quality-engineering-business-analysis` | 1025   | ███████░░░ 72%     | 10/10 | 6 | ✅ 95% |
| `quality-engineering-jira-integration` | 560    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `quality-engineering-quality-assurance` | 496    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `quality-engineering-zephyr-coverage-analysis` | 493    | █████████░ 87%     | 10/10 | 4 | ✅ 80% |
| `quality-engineering-zephyr-test-generation` | 1197   | ███████░░░ 67%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 react (8 skills | avg 521 tokens | quality 10.0/10 | eval alignment 96%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `react-component-patterns` | 458    | █████████░ 87%     | 10/10 | 3 | ✅ 100% |
| `react-hooks          ` | 623    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `react-performance    ` | 715    | ████████░░ 80%     | 10/10 | 3 | ✅ 100% |
| `react-security       ` | 491    | █████████░ 87%     | 10/10 | 3 | ✅ 83% |
| `react-state-management` | 523    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `react-testing        ` | 517    | █████████░ 86%     | 10/10 | 3 | ✅ 83% |
| `react-tooling        ` | 403    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `react-typescript     ` | 437    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 react-native (13 skills | avg 425 tokens | quality 10.0/10 | eval alignment 97%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `react-native-architecture` | 530    | █████████░ 86%     | 10/10 | 3 | ✅ 83% |
| `react-native-components` | 362    | █████████░ 90%     | 10/10 | 3 | ✅ 100% |
| `react-native-deployment` | 522    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `react-native-dls     ` | 256    | █████████░ 93%     | 10/10 | 3 | ✅ 100% |
| `react-native-navigation` | 325    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `react-native-navigation-v6` | 497    | █████████░ 86%     | 10/10 | 3 | ✅ 86% |
| `react-native-notifications` | 338    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `react-native-performance` | 563    | █████████░ 85%     | 10/10 | 3 | ✅ 89% |
| `react-native-platform-specific` | 395    | █████████░ 89%     | 10/10 | 3 | ✅ 100% |
| `react-native-security` | 565    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `react-native-state-management` | 425    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |
| `react-native-styling ` | 317    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `react-native-testing ` | 426    | █████████░ 88%     | 10/10 | 3 | ✅ 100% |

</details>

<details>
<summary><h3>📦 spring-boot (10 skills | avg 464 tokens | quality 10.0/10 | eval alignment 98%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `spring-boot-api-design` | 319    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-architecture` | 615    | ████████░░ 83%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-best-practices` | 564    | █████████░ 85%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-data-access` | 511    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-deployment` | 497    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-microservices` | 482    | █████████░ 87%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-observability` | 481    | █████████░ 87%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-scheduling` | 340    | █████████░ 91%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-security ` | 510    | █████████░ 86%     | 10/10 | 3 | ✅ 100% |
| `spring-boot-testing  ` | 317    | █████████░ 91%     | 10/10 | 3 | ✅ 83% |

</details>

<details>
<summary><h3>📦 swift (8 skills | avg 472 tokens | quality 10.0/10 | eval alignment 89%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `swift-best-practices ` | 649    | ████████░░ 82%     | 10/10 | 4 | ✅ 92% |
| `swift-concurrency    ` | 517    | █████████░ 86%     | 10/10 | 5 | ✅ 93% |
| `swift-error-handling ` | 500    | █████████░ 86%     | 10/10 | 4 | ✅ 100% |
| `swift-language       ` | 458    | █████████░ 87%     | 10/10 | 5 | ✅ 94% |
| `swift-memory-management` | 377    | █████████░ 90%     | 10/10 | 4 | ⚠️ 56% |
| `swift-swiftui        ` | 418    | █████████░ 89%     | 10/10 | 4 | ✅ 75% |
| `swift-testing        ` | 448    | █████████░ 88%     | 10/10 | 4 | ✅ 100% |
| `swift-tooling        ` | 411    | █████████░ 89%     | 10/10 | 4 | ✅ 100% |

</details>

<details>
<summary><h3>📦 typescript (4 skills | avg 725 tokens | quality 9.5/10 | eval alignment 100%)</h3></summary>

| Skill                   | Tokens | Savings (vs Heavy) | Quality | Evals | Aligned |
| ----------------------- | ------ | ------------------ | ------- | ----- | ------- |
| `typescript-best-practices` | 593    | ████████░░ 84%     | 10/10 | 3 | ✅ 100% |
| `typescript-language  ` | 648    | ████████░░ 82%     | 10/10 | 3 | ✅ 100% |
| `typescript-security  ` | 746    | ████████░░ 80%     | 10/10 | 5 | ✅ 100% |
| `typescript-tooling   ` | 912    | ████████░░ 75%     | 8/10 | 3 | ✅ 100% |

</details>

## ⚠️ Low Eval Alignment — Skills to Review

> These skills have evals but SKILL.md content does not cover ≥70% of what the evals test. The skill may not actually improve agent behavior for its target scenarios.

| Skill                   | Category | Alignment | Evals | Action |
| ----------------------- | -------- | --------- | ----- | ------ |
| `common-store-changelog` | common   | ⚠️ 43% | 4 | Add missing terms from eval assertions to SKILL.md |
| `flutter-security     ` | flutter  | ⚠️ 50% | 3 | Add missing terms from eval assertions to SKILL.md |
| `swift-memory-management` | swift    | ⚠️ 56% | 4 | Add missing terms from eval assertions to SKILL.md |

## 🏆 Quality Leaders

| Rank | Skill                   | Category | Quality | Tokens | Evals | Aligned |
| ---- | ----------------------- | -------- | ------- | ------ | ----- | ------- |
| 1    | `android-architecture ` | android  | 10/10 | 504 | 3 | ✅ 88% |
| 2    | `android-background-work` | android  | 10/10 | 302 | 3 | ✅ 100% |
| 3    | `android-compose      ` | android  | 10/10 | 447 | 3 | ✅ 100% |
| 4    | `android-concurrency  ` | android  | 10/10 | 314 | 3 | ✅ 89% |
| 5    | `android-deployment   ` | android  | 10/10 | 328 | 3 | ✅ 100% |
| 6    | `android-design-system` | android  | 10/10 | 252 | 3 | ✅ 100% |
| 7    | `android-di           ` | android  | 10/10 | 309 | 3 | ✅ 88% |
| 8    | `android-legacy-navigation` | android  | 10/10 | 310 | 3 | ✅ 86% |
| 9    | `android-legacy-security` | android  | 10/10 | 444 | 3 | ✅ 100% |
| 10   | `android-legacy-state ` | android  | 10/10 | 256 | 3 | ✅ 100% |

## 📐 Methodology & Baseline Justification

### Why These Baselines?

The baselines are derived from **real, token-counted example prompts** that represent what a developer actually writes when there is no structured skill available.

Using NestJS as the **Reference Unit**: Because we measure instruction volume replaced, using a high-density reference ensures scientific consistency across all tech stacks.

#### 🟡 Reference Technical Prompt — Light — 1449 tokens

> **Reference Technical Prompt — Light (e.g., NestJS)**
> A compact inline system prompt used as a reference for token count calibration. Representative of focused developer instructions without a structured skill.

#### 🔴 Reference Technical Prompt — Heavy — 3656 tokens

> **Reference Technical Prompt — Heavy (e.g., NestJS Architecture)**
> A comprehensive architect-level inline prompt used as a reference for complex tasks. Includes deep patterns and rules sent by developers when no skill is present.

### 🏆 Detailed Quality Rubric (0–10)

To ensure skills are not just "short" but actually **high quality**, every skill is scored against this structural rubric:

| Score  | Criteria                  | Rationale                                              |
| ------ | ------------------------- | ------------------------------------------------------ |
| **+2** | **Structured Guidelines** | At least 3 specific instructions/bullet points.                    |
| **+2** | **Anti-Patterns**         | `## Anti-Patterns` section or `**No X**` inline lines.            |
| **+2** | **Reference Examples**    | Presence of a verified `references/` folder with code.             |
| **+2** | **Token Optimality**      | Entire `SKILL.md` is ≤100 lines (forces brevity).                  |
| **+2** | **Eval Coverage**         | ≥3 evals with `should_not_trigger`, ≥2 assertions each. +1 partial.|

> **Eval Alignment** (reported separately, not scored): % of eval `contains` assertion values that appear in SKILL.md content. Measures whether the skill actually teaches what its evals test — the closest static proxy for **with-skill vs without-skill** behavioral improvement.

### 🛡️ How to Verify This Report

Trust but verify. You can audit the raw data and run the benchmark yourself:

1. **Clone the repo** and install dependencies (`pnpm install`).
2. **Inspect Source**: The benchmark logic is open in [cli/src/scripts/benchmark/](./cli/src/scripts/benchmark/).

### Pricing (per 1M input tokens, Feb 2026)

- **Gemini 3 Flash**: $0.50
- **GPT-5**: $1.25
- **Gemini 3.1 Pro**: $2.00
- **Claude Sonnet 4.5**: $3.00