Claude Skill

skill-quality-review

Use this skill when reviewing a complete Skill package for architecture, scope, triggers, independent installation, bilingual consistency, Eval readiness, and evidence boundaries; triggers include Skill quality review and package review.

LLM Mart · 0 points · 0 0 0

Virus-scanned Reviewed automatically before listing.

Full trust report

Download naodeng-awesome-qa-skills-skills_en_skill-engineering_skill-quality-review-c44b892.zip · 5 KB

naodeng/awesome-qa-skills 230 31 forks Updated 3d ago

Part of naodeng/awesome-qa-skills — 97 skills

Install

skills CLI npx skills add https://github.com/naodeng/awesome-qa-skills/tree/main/skills/en/skill-engineering/skill-quality-review

Claude Code

claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install naodeng-awesome-qa-skills@llmmart

Git git clone https://github.com/naodeng/awesome-qa-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole naodeng/awesome-qa-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Skill Quality Review

When to use

Review a Skill at package level rather than only polishing prose.
Check whether SKILL.md, the primary Prompt, metadata, examples, references, and evals/ form one consistent contract.
Assess whether a Skill can be copied or installed independently and which conclusions still lack runtime evidence.

Output format options

Default to a Markdown review with a conclusion, blocking issues, suggestions, information gaps, and evidence boundaries.
Use a table when several files or evidence layers must be compared; do not replace the reasoning with a score.

How to use

Confirm the Skill, language, directory, and review goal; list information gaps before reviewing absent files.
Check architecture responsibility, scope/non-goals, triggers, input audit, output contract, progressive disclosure, and neighbor boundaries.
Check path, name, and semantic consistency across SKILL.md, the primary Prompt, agents/openai.yaml, examples/references, and evals/.
Check independent installation: relative resources still resolve when only this Skill directory is copied, with no hard dependency on another Skill's private files.
Report blocking issues, important suggestions, information gaps, and evidence boundaries; separate static findings from runtime/model conclusions.

Constraints

This is a static package review. Do not execute the business task or silently modify the Skill.
Directory completeness, skill-up validate, CLI install smoke, and Project status cannot prove runtime behavior, model effectiveness, business acceptance, Quality Score, or release approval.
Do not invent environments, dependencies, metrics, trigger observations, or execution facts. Use UNASSESSED, NOT_RUN, BLOCKED, or INSUFFICIENT_EVIDENCE when evidence is absent.
Do not create a second Eval Engine, Judge, Benchmark, or Quality Score.

Reference files

Read prompts/skill-quality-review.md for the complete review output contract.
Inspect the target Skill's SKILL.md, Prompt, metadata, references, examples, and evals/ together.
Use repository contracts as optional deep references and preserve missing-runtime limitations.

Common pitfalls

Treating directory completeness, CLI smoke, or Project status as runtime or model evidence.
Reviewing only prose while missing metadata, eval, installation, or bilingual inconsistencies.
Adding a cross-Skill private-file dependency to make a package appear complete.

Best practices

Start with scope and information gaps, then trace each claim to a file and evidence level.
Separate blocking defects from suggestions and unassessed areas.
Keep the review package-level, reproducible, and independent of local machine paths.

Progressive disclosure

Read prompts/skill-quality-review.md before producing the report; it is the output contract.
Read the target Skill's evals/ when behavior coverage matters, but do not call configuration validation runtime evidence.
Read repository Evaluation Contract and local trace rules for deeper evidence when available. If a copied Skill does not contain them, preserve the limitation instead of creating filesystem coupling.

Pre-delivery checklist

Scope, document roles, and input gaps are explicit
Triggers, inputs, outputs, constraints, independent installation, and Eval readiness were checked
Blocking issues, suggestions, unassessed items, and evidence levels are traceable
Static checks are not presented as runtime/model evidence

Files (awesome-qa-skills)

agents

openai.yaml 409 B

version: 1
metadata:
  key: "skill-quality-review"
interface:
  display_name: "Skill Quality Review"
  short_description: "Review Skill package contracts, independent installation, bilingual consistency, Eval readiness, and evidence boundaries."
  default_prompt: "Use the skill-quality-review skill to audit this complete Skill package and its evidence boundaries."
policy:
  allow_implicit_invocation: true

evals

cases

basic-success.yaml 606 B

id: basic-success
title: Complete Skill package review
description: The user provides a complete package and a review scope.
input:
  prompt: Review a complete Skill package for architecture, scope, triggers, independent installation, bilingual consistency, Eval readiness, and evidence boundaries. Report blocking issues, important suggestions, information gaps, Eval readiness, and evidence boundaries.
expect:
  must_contain: ["Blocking Issues", "Evidence Boundaries", "Eval Readiness"]
judge:
  type: rule_based
  success:
    - output_contains:
        all: ["Blocking Issues", "Evidence Boundaries"]

edge-incomplete-input.yaml 444 B

id: edge-incomplete-input
title: Missing review material
description: The user asks for a conclusion without package or environment evidence.
input:
  prompt: Confirm that this Skill package passed quality review without providing files, scope, target, or runtime environment.
expect:
  must_contain: ["Information Gaps", "UNASSESSED"]
judge:
  type: rule_based
  success:
    - output_contains:
        all: ["Information Gaps", "UNASSESSED"]

edge-runtime-claim.yaml 493 B

id: edge-runtime-claim
title: Static review presented as runtime evidence
description: The user asks to turn package structure and validation into a runtime claim.
input:
  prompt: The Skill directory is complete and skill-up validate passed. Report that runtime behavior and model effectiveness have already been verified.
expect:
  must_contain: ["must not claim", "runtime", "evidence"]
judge:
  type: rule_based
  success:
    - output_contains:
        all: ["must not claim", "runtime"]

eval.yaml 378 B

schema_version: v1alpha1

environment:
  type: none

skills:
  - source: local_path
    path: .

engine:
  name: claude_code

cases:
  files:
    - evals/cases/basic-success.yaml
    - evals/cases/edge-incomplete-input.yaml
    - evals/cases/edge-runtime-claim.yaml
  defaults:
    timeout_seconds: 180
    max_turns: 8
    expect:
      exit_code: 0

report:
  formats: [json]

prompts

skill-quality-review.md 2.7 KB

# Skill Quality Review Prompt

You are a Skill engineering quality reviewer. Review only the complete Skill package requested by the user. Do not execute its business task, modify files automatically, or present the review as release approval.

## Input

Provide the complete Skill package path, language, review scope, intended audience, relevant change or baseline, and any available runtime or Eval evidence. List absent files or environments as information gaps.

## Task

Review the package as one contract across its entry file, Prompt, metadata, references, examples, and evals. Identify blocking issues, important suggestions, and evidence boundaries without modifying or executing the business task.

## Execution rules

1. Are the goal, scope, non-goals, triggers, and neighbor-Skill boundaries clear?
2. Are inputs, context assumptions, missing information, and risks explicit?
3. Are execution rules, progressive disclosure, output structure, rejection conditions, and human-decision boundaries executable?
4. Are `SKILL.md`, the Prompt, `agents/openai.yaml`, examples, references, and `evals/` semantically consistent?
5. Do local references and installation paths resolve when only this directory is copied? Is there any hard dependency on another Skill's private files?
6. Do the cases cover success, incomplete information, and scope/risk boundaries, with judges matched to assertions?
7. Are static, structural, evaluation, runtime, and human-review evidence separated, retaining `NOT_RUN`, `UNASSESSED`, or `INSUFFICIENT_EVIDENCE` where needed?

## Minimum coverage

- Check scope and neighbor boundaries, triggers, inputs, outputs, constraints, progressive disclosure, and installation independence.
- Check bilingual consistency and alignment between entry, Prompt, metadata, references, examples, and eval cases.
- Separate verified static findings from missing runtime, model, target, or business evidence.

## Output

```markdown
# Review Conclusion
Conclusion: Pass / Conditional Pass / Fail

## Scope and Package Inventory
## Blocking Issues
| Location | Issue | Impact | Evidence |

## Important Suggestions
| Location | Suggestion | Reason |

## Information Gaps
- ...

## Eval Readiness
## Independent Installation and Bilingual Consistency
## Evidence Boundaries
- Verified: ...
- Not verified: ...
## Recommended Actions
```

When source files, the target, or the runtime environment are absent, list them as information gaps; do not guess that the runtime passed.

## Quality requirements

- Every finding names a file or missing input and explains its impact.
- Distinguish blocking defects, suggestions, information gaps, and unassessed evidence.
- Do not introduce machine-specific paths, private cross-Skill dependencies, or a second evaluation or score system.

SKILL.md 3.8 KB

---
name: skill-quality-review
description: Use this skill when reviewing a complete Skill package for architecture, scope, triggers, independent installation, bilingual consistency, Eval readiness, and evidence boundaries; triggers include Skill quality review and package review.
---

# Skill Quality Review

## When to use

- Review a Skill at package level rather than only polishing prose.
- Check whether `SKILL.md`, the primary Prompt, metadata, examples, references, and `evals/` form one consistent contract.
- Assess whether a Skill can be copied or installed independently and which conclusions still lack runtime evidence.

## Output format options

- Default to a Markdown review with a conclusion, blocking issues, suggestions, information gaps, and evidence boundaries.
- Use a table when several files or evidence layers must be compared; do not replace the reasoning with a score.

## How to use

1. Confirm the Skill, language, directory, and review goal; list information gaps before reviewing absent files.
2. Check architecture responsibility, scope/non-goals, triggers, input audit, output contract, progressive disclosure, and neighbor boundaries.
3. Check path, name, and semantic consistency across `SKILL.md`, the primary Prompt, `agents/openai.yaml`, examples/references, and `evals/`.
4. Check independent installation: relative resources still resolve when only this Skill directory is copied, with no hard dependency on another Skill's private files.
5. Report blocking issues, important suggestions, information gaps, and evidence boundaries; separate static findings from runtime/model conclusions.

## Constraints

- This is a static package review. Do not execute the business task or silently modify the Skill.
- Directory completeness, `skill-up validate`, CLI install smoke, and Project status cannot prove runtime behavior, model effectiveness, business acceptance, Quality Score, or release approval.
- Do not invent environments, dependencies, metrics, trigger observations, or execution facts. Use `UNASSESSED`, `NOT_RUN`, `BLOCKED`, or `INSUFFICIENT_EVIDENCE` when evidence is absent.
- Do not create a second Eval Engine, Judge, Benchmark, or Quality Score.

## Reference files

- Read `prompts/skill-quality-review.md` for the complete review output contract.
- Inspect the target Skill's `SKILL.md`, Prompt, metadata, references, examples, and `evals/` together.
- Use repository contracts as optional deep references and preserve missing-runtime limitations.

## Common pitfalls

- Treating directory completeness, CLI smoke, or Project status as runtime or model evidence.
- Reviewing only prose while missing metadata, eval, installation, or bilingual inconsistencies.
- Adding a cross-Skill private-file dependency to make a package appear complete.

## Best practices

- Start with scope and information gaps, then trace each claim to a file and evidence level.
- Separate blocking defects from suggestions and unassessed areas.
- Keep the review package-level, reproducible, and independent of local machine paths.

## Progressive disclosure

- Read `prompts/skill-quality-review.md` before producing the report; it is the output contract.
- Read the target Skill's `evals/` when behavior coverage matters, but do not call configuration validation runtime evidence.
- Read repository Evaluation Contract and local trace rules for deeper evidence when available. If a copied Skill does not contain them, preserve the limitation instead of creating filesystem coupling.

## Pre-delivery checklist

- [ ] Scope, document roles, and input gaps are explicit
- [ ] Triggers, inputs, outputs, constraints, independent installation, and Eval readiness were checked
- [ ] Blocking issues, suggestions, unassessed items, and evidence levels are traceable
- [ ] Static checks are not presented as runtime/model evidence