agent-evaluation
Testing and benchmarking LLM agents including behavioral testing, capability assessment, reliability metrics, and production monitoring—where even top agents achieve less than 50% on real-world benchmarks
42
30%
Does it follow best practices?
Impact
Pending
No eval scenarios have been run
Passed
No known issues
Optimize this skill with Tessl
npx tessl skill review --optimize ./skills/antigravity-agent-evaluation/SKILL.mdQuality
Discovery
32%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
The description identifies a clear domain (LLM agent testing and benchmarking) and lists relevant subcategories, but it reads more like a topic summary than an actionable skill description. It lacks a 'Use when...' clause, concrete actions the skill performs, and the trailing statistical claim ('even top agents achieve less than 50%') is editorial commentary that doesn't aid skill selection.
Suggestions
Add an explicit 'Use when...' clause with trigger scenarios, e.g., 'Use when the user asks about evaluating LLM agents, creating test suites for AI systems, measuring agent reliability, or setting up evals.'
Replace category labels with concrete actions, e.g., 'Designs behavioral test suites, runs capability benchmarks, calculates reliability metrics, and sets up production monitoring dashboards for LLM agents.'
Remove the editorial claim about '50% on real-world benchmarks' and instead add common user trigger terms like 'eval', 'evaluation', 'agent performance', 'test harness', or 'accuracy measurement'.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Names the domain (LLM agent testing/benchmarking) and lists some action areas (behavioral testing, capability assessment, reliability metrics, production monitoring), but these are more like categories than concrete actions. It doesn't specify what the skill actually does (e.g., 'generates test suites', 'runs benchmarks', 'produces reports'). | 2 / 3 |
Completeness | Describes what the skill covers at a high level but completely lacks a 'Use when...' clause or any explicit trigger guidance for when Claude should select this skill. The rubric states a missing 'Use when...' clause should cap completeness at 2, and the 'what' itself is also somewhat vague, bringing this to 1. | 1 / 3 |
Trigger Term Quality | Includes relevant terms like 'LLM agents', 'benchmarking', 'testing', 'reliability metrics', and 'production monitoring' which users might naturally use. However, it misses common variations like 'eval', 'evaluation', 'agent evaluation', 'test harness', 'accuracy', or 'performance testing'. | 2 / 3 |
Distinctiveness Conflict Risk | The focus on LLM agent testing/benchmarking is a reasonably specific niche, but terms like 'testing', 'monitoring', and 'metrics' could overlap with general software testing or monitoring skills. The statistical claim about '50% on real-world benchmarks' adds flavor but doesn't help with disambiguation. | 2 / 3 |
Total | 7 / 12 Passed |
Implementation
27%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is comprehensive in coverage but severely undermined by its verbosity and monolithic structure. The ~700+ lines of mostly non-executable TypeScript class implementations could be condensed to a fraction of the size by providing concise patterns with key interfaces and delegating full implementations to separate reference files. The content reads more like a tutorial or library documentation than a lean skill file optimized for Claude's context window.
Suggestions
Reduce the main file to ~100 lines with concise pattern summaries and key interfaces, moving full class implementations to separate reference files (e.g., patterns/statistical-evaluation.md, patterns/adversarial-testing.md)
Remove metadata sections (Capabilities, Prerequisites, Scope, Ecosystem, Related Skills, When to Use) that duplicate frontmatter concerns and don't provide actionable guidance
Make code examples executable by either providing concrete implementations with real dependencies (e.g., using actual testing frameworks like Jest/Vitest) or simplifying to focused, runnable snippets rather than abstract class hierarchies
Add explicit validation checkpoints to workflows, e.g., 'If baseline pass rate < 0.7, stop and investigate test quality before proceeding to regression testing'
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Extremely verbose at ~700+ lines. Massive code blocks contain full class implementations with helper methods, type definitions, and extensive inline comments that Claude doesn't need. Concepts like confidence intervals, chi-squared tests, and Jaccard similarity are explained through code that Claude already understands. The metadata sections (Capabilities, Prerequisites, Scope, Ecosystem) add significant overhead with information that could be drastically condensed. | 1 / 3 |
Actionability | The code examples are detailed TypeScript/pseudo-TypeScript but not truly executable—they reference undefined types (Agent, AgentOutput, AgentContext, TestCase), unimplemented helper methods (containsRudeLanguage, isRelevantToCustomerService, containsLegalAdvice, similarity), and abstract interfaces without concrete implementations. They illustrate patterns well but aren't copy-paste ready. | 2 / 3 |
Workflow Clarity | The Collaboration section includes brief workflow sequences (design → create suite → implement → evaluate → iterate), and the patterns are logically sequenced. However, there are no explicit validation checkpoints within the main patterns—the statistical evaluator and regression tester don't specify when to stop, what thresholds trigger action, or how to recover from failures in the evaluation pipeline itself. | 2 / 3 |
Progressive Disclosure | This is a monolithic wall of text with no references to external files. All patterns, sharp edges, and collaboration workflows are inlined in a single massive document. The content would benefit enormously from splitting patterns into separate files (e.g., STATISTICAL_TESTING.md, ADVERSARIAL_TESTING.md, REGRESSION_TESTING.md) with a concise overview in the main skill file. | 1 / 3 |
Total | 6 / 12 Passed |
Validation
81%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 9 / 11 Passed
Validation for skill structure
| Criteria | Description | Result |
|---|---|---|
skill_md_line_count | SKILL.md is long (1132 lines); consider splitting into references/ and linking | Warning |
frontmatter_unknown_keys | Unknown frontmatter key(s) found; consider removing or moving to metadata | Warning |
Total | 9 / 11 Passed | |
- Repository
- boisenoise/skills-collections
- Commit
636b862
- Reviewed
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.