eval-harness

适用于 Claude Code 会话的正规评测框架（Evaluation Framework），实现了评测驱动开发（Eval-Driven Development, EDD）原则

Quality

57%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./.agents/skills/eval-harness/SKILL.md

Quality

Content

65%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The content is highly actionable with concrete, executable grader and command templates and a clear workflow skeleton. Its weaknesses are a lack of explicit validate-and-retry feedback loops in the workflow and a monolithic single-file structure with no progressive disclosure.

Suggestions

Add an explicit validation/feedback checkpoint in the evaluate step, e.g. "if any eval FAILS, fix the code and re-run evals before generating the report," to lift workflow_clarity.

Split large reference material (grader templates, pass@k/pass^k metrics, the full auth example) into a separate reference file with clearly signaled one-level-deep links to improve progressive_disclosure.

Trim the philosophy framing and condense the worked auth example to make the body leaner and improve conciseness.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient and assumes Claude's competence without basic-concept padding, but the philosophy framing and the lengthy worked auth example add content that could be trimmed; not maximally lean, so it is a 2 rather than a 3.	2 / 3
Actionability	It provides executable bash graders (grep -q, npm test, npm run build), concrete slash commands (/eval define\|check\|report), and copy-paste-ready eval/report templates, fully meeting the executable-and-specific anchor.	3 / 3
Workflow Clarity	The define→implement→evaluate→report sequence is clearly listed, but validation feedback loops are implicit (evaluate records PASS/FAIL then jumps to report) with no explicit fix-and-re-eval-before-reporting checkpoint, capping it at 2.	2 / 3
Progressive Disclosure	No bundle files exist and all content is inline in one ~228-line monolithic file; sections are well-organized but content that could be split (grader templates, metrics, worked examples) stays inline with no one-level-deep references, so it is a 2 rather than a 3.	2 / 3
	Total	9 / 12 Passed

Description

50%Weight 40%Scale 1-3

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 clearly names its domain (an evaluation framework implementing EDD) but stops short of listing concrete capabilities and omits any explicit "Use when..." trigger guidance. It is competent but incomplete for activation purposes.

Suggestions

Add concrete actions the skill performs, e.g. "define capability/regression evals, run code- and model-based graders, and track pass@k reliability".

Append an explicit "Use when..." clause naming natural user triggers (e.g. setting up eval-driven development, defining pass/fail criteria, creating regression suites, benchmarking agent performance) to lift completeness and trigger quality.

Include common natural phrasings users would say (e.g. "evals", "regression tests", "pass@k", "benchmark agent reliability") to improve trigger-term coverage.

Dimension	Reasoning	Score
Specificity	Names a concrete domain ("正规评测框架"/Evaluation Framework) and the EDD principle it implements, but does not list multiple specific concrete actions like defining evals or running graders, so it is not a 3; it is more concrete than a 1 like "helps with documents".	2 / 3
Completeness	It states what the skill is, but there is no "Use when..." or equivalent explicit trigger guidance describing when Claude should use it; per the rubric a missing when-clause caps completeness at 2.	2 / 3
Trigger Term Quality	Relevant domain keywords ("评测框架", "评测驱动开发", "EDD") are present but they lean technical and omit common natural user phrasings, so coverage is partial rather than a 1 (pure jargon) or a 3 (full natural-term coverage).	2 / 3
Distinctiveness Conflict Risk	It targets a fairly specific niche (EDD for Claude Code sessions), but the absence of explicit triggers means it is not clearly distinguishable by trigger alone, leaving it somewhat specific but still potentially overlapping rather than a clear 3.	2 / 3
	Total	8 / 12 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: xu-xiang/everything-claude-code-zh
Path: .agents/skills/eval-harness/SKILL.md
Commit: dfbf946

Reviewed: 28 days ago

Table of Contents

Discovery Implementation Validation

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.