Receber por email
← Voltar para o catálogo

page-report

Shared report-page contract for Agentic SEO modules that render editable, human-first Markdown reports in the Web Companion.

45estrelas
Atualizado há 5 meses

Ver no GitHub ↗ Licença: MIT

Como adicionar

/plugin marketplace add agencia-conversion/agentic-seo-skills

O comando exato pode variar conforme o repositório. Confira o README no GitHub.

Para o autor da skill

Cole no README do seu repo

Mostra que sua skill está catalogada na Skillteca, gera backlink e tráfego rastreável.

[![Skillteca](https://www.skillteca.com.br/api/badge/page-report/svg)](https://www.skillteca.com.br/skills/page-report)

Skills relacionadas

Ver todas de Dados e Análise →

Alerta por categoria

Receba novas skills de Dados e Análise toda segunda

1 email curto, só com as skills novas de Dados e Análise. 4 minutos de leitura, sem spam, cancela com 1 clique.

Você confirma o email no primeiro envio. Sem spam. Cancela com 1 clique.

Ler descrição completa

Page Report

Use this skill whenever an Agentic SEO workflow generates a report page for the Web Companion. It is a shared contract, not a standalone SEO module.

When To Use

Use page-report from report-producing skills and runtime code when creating or reviewing report.md for Technical SEO, Internal Links, SEO Analysis, Keyword Research, SERP Extract, Backlink Analysis, Topic Cluster, E-E-A-T, or Competitive Analysis.

Do not use this skill for public content drafts, approval handoffs, inline chat answers, raw provider captures, or authorial Brain updates that are not report pages.

Critical Points

  • The canonical report path is project/analyses/<module>/<run-slug>/report.md.
  • The report is the editable human presentation layer. Raw evidence, normalized YAML/JSON, logs, and operational artifacts stay in sources/, audits/, keywords/, clusters/, eeat/, or workbench/.
  • Frontmatter must include title, slug, report_type, generated_at, status, source_artifact, and summary; include score only when the module has a real score.
  • Do not write # <title> in the Markdown body. The Companion renders the title from frontmatter.
  • Use the project language from project/.agentic-seo/project.json.language unless a command has an explicit language override. Complete copy is required for pt-BR and en.
  • New visual fences must be YAML with version: 1: agentic-kpis, agentic-chart, and agentic-table. JSON fences are legacy read compatibility only.
  • Tables must use stable lower-snake ASCII key values. User-facing label values may be translated or edited without changing keys.
  • Calculation tables must mark columns with role: weight, role: points, and role: loss so score recalculation survives label edits.
  • Each report-producing module has skills/<module>/contract.yaml and templates/analyses/<module>/report-skeleton.md; keep required H2 sections and table keys from the skeleton.
  • Before declaring a manual report done, run npm run validate:reports -- --module=<module> --slug=<run-slug> or use scripts/lib/report-writer.mjs, which refuses invalid reports before writing.
  • Never paste raw provider JSON, object arrays, [object Object], internal IDs, or snake_case evidence into the visual body. Convert evidence into natural language and readable tables.
  • Format numbers per the project language. The canonical source is shared/locale.mjs (getProjectLanguage, formatNumber, formatPercent, canonicalKeyword). Skills either ship locale-formatted strings in agentic-table/agentic-kpis cells, or leave numbers raw for the Companion (useI18n().formatNumber / formatPercent) to format on render. Percent-bearing columns use the _pct key suffix on the 0..100 scale.
  • Always return report_md and browser_prompt: { recommended: true, message: "Posso abrir o Web Companion para você ver a análise?", open_with: "project-browser" }.
  • Register the generated report in project/brain/log.md as type: decision unless a more specific log type is required.

Framework

1. Build The Source Artifact

Write the module's authoritative machine-readable artifact first. It may be YAML or JSON, but it must remain outside project/analyses/ and be referenced by frontmatter source_artifact.

2. Render The Human Page

Start with an executive reading. Then add module-specific sections, visual blocks, and human-readable appendices. Keep technical depth, but translate it into prose, friendly labels, and compact evidence.

3. Validate The Contract

Before returning the result, verify: canonical path, frontmatter, no H1 body title, no HTML default, YAML visual fences, no raw JSON/object dumps, readable evidence, source separation, and log entry.

4. Review The Result

For QA, generate one report per module with fixtures/offline data. A report is satisfactory only when it scores at least 4/5 for clarity, evidence separation, visual usefulness, actionability, and localization, with no category below 3/5.

Output Format

The report Markdown must have this shape:

---
title: "Human report title"
slug: "stable-run-slug"
report_type: "technical-seo"
generated_at: "2026-05-23T00:00:00.000Z"
status: "ready"
source_artifact: "audits/example/report.yaml"
summary: "One-sentence executive summary."
score: "92"
---

Executive summary prose without a duplicate H1.

version: 1 items:

  • label: Score

value: 92/100

Use only supported report_type values: technical-seo, internal-links, seo-analysis, keyword-research, serp-extract, backlink-analysis, topic-cluster, eeat, and competitive-analysis.

Done Criteria

  • report.md exists under the canonical module path and is returned as report_md.
  • source_artifact points to the normalized technical source.
  • The Companion can list, open, edit, save, and roundtrip the page.
  • Visual fences parse as YAML version: 1.
  • No visible JSON/object dumps or duplicated title appear in the report body.
  • The report is registered in brain/log.md.
Compartilhar X LinkedIn