Skip to main content

review-docs

Review Apify documentation for style guide compliance, quality standards, and best practices. Use when user says "review this doc", "check this page", "audit documentation", "review before PR", "is this ready to publish", or "review-docs". Runs automated checks and manual review against Apify style guide.

Informações da origem

Repositório
apify/apify-docs
Última atividade na origem
15 de setembro de 2026 às 06:50
Idioma detectado do SKILL.md
inglês
Estrelas
81
Forks
239

Opções de instalação

Por padrão, está selecionado o prompt que primeiro revisa a origem. Você pode mudar para um comando direto ou baixar uma cópia local.

Revise os arquivos de origem

Leia o SKILL.md e os arquivos complementares exibidos pelo SkillsMP antes de decidir se vai instalar.

Explorador de arquivos
4 arquivos

Exibindo SKILL.md

SKILL.md
Instruções da origem · Visualização somente leitura
name
review-docs
description
Review Apify documentation for style guide compliance, quality standards, and best practices. Use when user says "review this doc", "check this page", "audit documentation", "review before PR", "is this ready to publish", or "review-docs". Runs automated checks and manual review against Apify style guide.
allowed-tools
Read Bash Glob Grep Agent
model
sonnet
argument-hint
file-path
# Documentation review ## Process 1. **Verify file version** - `git status` to confirm you have the latest 2. **Run deterministic checks** (main process) - these are objective, no judgment needed: - `vale --minAlertLevel=suggestion "<file>"` (prose style: voice, tone, terminology, grammar, headings, link text) - `pnpm lint:md` (heading hierarchy, list numbering, spacing) - `.agents/skills/review-docs/scripts/check-frontmatter.sh "<file>"` (description char count) Vale carries most of the style guide, so don't re-check its ground by hand. If it isn't installed, say so and note that prose coverage was skipped. For a pull request, run all three over every changed `.md` and `.mdx` file and report per file. 3. **Delegated review** - spawn two subagents in parallel, for what no tool can check. Each reads the page being reviewed plus one standards file, and returns findings with line numbers and suggested fixes: - Subagent 1, `standards/style-guide.md`: bold outside UI elements and critical warnings, link text that doesn't say where it goes, non-parallel lists, missing serial commas, and loose use of "legacy", "alternative", or "deprecated" - Subagent 2, `standards/page-structure.md`: stacked headings, concepts used before they're explained, screenshots that don't earn their place or break the treatment rules, admonitions that don't fit their content, outnumber one per H2 section, or carry content the prose needs 4. **Content review** (main process) - focus on what neither tools nor standards cover: - Content structure (clear intro, logical flow, next steps) - Technical accuracy (code examples correct, API endpoints current) - Completeness (prerequisites listed, edge cases covered) - Code example quality (complete, runnable, commented where needed) 5. **Format output** - per `.agents/skills/review-docs/references/review-format.md`. Tool output goes in its own section, verbatim with rule names and line numbers. Judgment findings go in a separate section and must not restate anything a tool reported. For detailed process notes and edge cases, see `.agents/skills/review-docs/references/process.md`.
Ver no GitHub