| name | quality-report |
| description | Generate a formal data quality coverage report in the format chosen by the user — chat, PDF, DOCX, PPTX, web dashboard, web article, poster or XLSX coverage matrix. Delegates file generation to the agent's writer skills (pdf-writer, docx-writer, pptx-writer, web-craft, canvas-craft, xlsx-writer) plus the centralized theming skill; chat format works standalone, file formats are offered only when the host agent declares the matching writer skill. Use when the user wants a quality report, coverage summary, quality dashboard, executive deck of quality findings, or any deliverable that consolidates assess-quality results. |
| argument-hint | [format: chat|pdf|docx|pptx|web|poster|xlsx] [filename (optional)] |
Skill: Quality Report Generation
Guidance-first workflow to produce a structured data quality coverage report. This skill orchestrates: it collects and composes the content, resolves the brand theme, and delegates the file-format generation to the host agent's writer skills. Only the Chat format is produced standalone by this skill.
1. Prerequisites and report data
This skill needs quality data to generate the report. Check if it already exists in the current conversation:
If coverage assessment or rule creation data already exists in the conversation: use that data directly. This includes both rules created from the gap flow (Flow A) and specific rules created directly by the user (Flow B).
If there is NO coverage data in the context (rule inventory, gaps, EDA): a full assessment of the requested scope must be performed first before generating the report. Inform the user and stop.
Data to collect for the report
If the data is already in context, extract it directly. If missing, obtain it with parallel MCP calls:
Parallel:
A. get_tables_quality_details(domain_name, tables)
B. get_table_columns_details(domain_name, table) [for each table]
C. get_quality_rule_dimensions(domain_name=domain_name)
D. get_critical_data_elements(domain_name=domain_name) [if not already in context; non-blocking — on permission/error, omit the CDE column from the report]
2. Format selection
2.1 Pre-check — writer skills availability (MANDATORY first step)
Before offering file-format options, confirm which writer skills the host agent declares. If the host agent does NOT declare pdf-writer, docx-writer, pptx-writer, web-craft, canvas-craft nor xlsx-writer, the only format available is Chat — skip §2.2 entirely and proceed with §5.1 (Chat).
This check is blocking: DO NOT offer file formats that the host agent cannot materialise, and DO NOT attempt to load a writer skill that is not declared by the host agent, even if the user insists on a file format.
2.2 Format options
If at least one writer skill is declared, ask the user following the user question convention. Only offer the options whose required skill is declared by the agent:
- Chat — structured summary in this conversation (no file). Always available.
- PDF — typographic multi-page document. Requires
pdf-writer.
- DOCX — editable Word document. Requires
docx-writer.
- PowerPoint — executive summary deck (16:9 by default). Requires
pptx-writer.
- Dashboard web — interactive HTML with filters, KPI cards and sortable tables. Requires
web-craft.
- Web article / Narrative report — self-contained HTML page, narrative or editorial layout. Requires
web-craft.
- Poster / Infographic — single-page visual summary for print or publication. Requires
canvas-craft.
- Excel (XLSX) — multi-sheet tabular coverage workbook (cover + coverage matrix with conditional formatting + rules detail + gaps + recommendations). Requires
xlsx-writer.
Multiple selection is allowed (the same content materialises in several formats using the same internal quality-report.md).
If the user specified the format in the arguments or message, use it directly and skip the question.
3. Report structure
The report has the same structure regardless of format. Six canonical sections in this fixed order:
- Cover / Header — title, domain, scope, generation date, agent name.
- Executive summary — tables analysed, rules total, breakdown OK/KO/WARNING/NOT_EXECUTED, coverage estimate, gaps by priority (CRITICO/ALTO/MEDIO/BAJO), rules created this session.
- Coverage by table — matrix of tables × dimensions with icon-coded status.
- Rules detail — per table: rules with name, dimension, status, % pass, description. KO and WARNING visually highlighted.
- Identified gaps — prioritised list with table, column, dimension, priority, description and recommendation.
- Recommendations and next steps — prioritised bullets.
If the user created rules via create-quality-rules in the current conversation, an optional Rules created in this session section goes between §4 and §5.
For the full layout contract (iconography, KPI cards, per-format composition, deterministic rules for audit), see quality-report-layout.md in this skill folder.
4. Branding decisions
Before invoking any writer skill for a file format, fix the theme following the host agent's branding cascade. When the host agent declares a format→skill contract with a §Branding decisions sub-section, follow that cascade — it is typically 4 levels: pin → explicit signal → intra-session continuity → curated proposal.
Primary neutral default for quality reports (when no cascade rule resolves to a specific theme): forensic-audit. It matches the audit register of a coverage report and keeps month-over-month reruns of the same dataset visually stable. The curated proposal should prefer themes whose descriptor mentions "audit", "technical", "editorial" or "corporate".
If the user asks for neutrality explicitly ("no me importa el diseño" / "make it neutral" / "hazlo neutro"), apply technical-minimal as the sober fallback — maximum restraint, predictable output for non-audit contexts.
The chosen theme is recorded silently as the last line of the internal quality-report.md (e.g. theme applied: forensic-audit). Informational, not a contract.
5. Deliverable generation
5.1 Chat format
Emit the report directly as markdown in the current response, following the six canonical sections from §3. No file is produced, no writer skill is invoked, the branding cascade does not run.
End the message with the summary of 2-3 key points per §7.
5.2 Markdown on disk (optional, trivial)
If the user asks for a .md file on disk (not Chat, not one of the writer-backed formats), write it directly with the Write tool at output/<folder>/<slug>-quality-report.md. No writer skill involved — markdown is plain text.
Use the same folder convention as §5.3.
5.3 File formats (PDF, DOCX, PPTX, Dashboard web, Web article, Poster)
-
Folder: create output/YYYY-MM-DD_HHMM_quality_<slug>/ (all artefacts live here). <slug> = domain or scope normalised (lowercase ASCII, accents stripped, spaces replaced by underscores, max 30 chars). Example: 2026-04-24_1530_quality_analiticabanca.
-
Brand tokens (once, before any visual format): load the centralized theming skill (brand-kit) with the theme fixed in §4. The theme file provides the token bundle — colours, typography, chart palette — that downstream writer skills consume so every deliverable looks coherent.
-
Assemble the content source: write output/<folder>/quality-report.md in the user's language, containing every canonical section from §3 using the structure described in quality-report-layout.md §10. When a section has no data, include the heading and an explicit "No X detected" line — do not omit silently. Apply the deterministic layout rules from quality-report-layout.md §11 (fixed section order, KPI card order, matrix column order, gap ordering, item limits per format). This internal file is the single source of truth the writer skills consume.
-
For each selected format, load the matching writer skill and produce the deliverable. Follow quality-report-layout.md for the format-specific composition:
- PDF → load
pdf-writer. Output: <slug>-quality-report.pdf. The deliverable must render the six canonical sections, apply brand tokens, honour the KPI cards pattern from the layout guide §5, and follow the PDF composition rules from §6.1 (header/footer on every page, repeat matrix header on page breaks, portrait unless dimension count >8).
- DOCX → load
docx-writer. Output: <slug>-quality-report.docx. Same six sections; use Word-native tables for the coverage matrix and rules detail so the user can edit them. Preserve heading styles (Heading 1, Heading 2, …).
- PowerPoint → load
pptx-writer. Output: <slug>-quality-summary.pptx. 16:9 by default; 4:3 only if the user asked explicitly. Target ≤12 slides following the outline in quality-report-layout.md §6.3.
- Dashboard web → load
web-craft. Output: <slug>-quality-dashboard.html. Apply the quality-specific layout from §6.4 (exactly the four KPI cards from §5, interactive coverage heatmap, filters for dimension/status/priority, sortable tables). If the host agent also declares a general analytical-dashboard.md guide, load it alongside so the generic dashboard patterns (global filters, Plotly via CDN, data budget, responsive, motion, language) apply on top.
- Web article / Narrative report → load
web-craft. Output: <slug>-quality-article.html. Do NOT apply the dashboard layout from §6.4. Use artifact class Page or Article. Render the six canonical sections as a scrollable, self-contained HTML page: narrative headings, inline KPI callouts, embedded Plotly coverage charts as figures, static HTML tables (not sortable). Same content as other formats; presentation is editorial, not interactive.
- Poster / Infographic → load
canvas-craft. Output: <slug>-quality-poster.pdf or <slug>-quality-poster.png. Single-page composition per §6.5 (KPI band top, central heatmap, top-3 gaps + top-3 recommendations in flanking columns, A3 portrait by default).
- Excel (XLSX) → load
xlsx-writer. Output: <slug>-quality-report.xlsx. Multi-sheet workbook per quality-report-layout.md §6.6 (Cover with 4 KPI cards as merged-cell band, Coverage matrix as native Table with conditional formatting for OK/KO/WARNING/NOT_EXECUTED, Rules detail grouped by table, Gaps prioritised, optional Rules-created-this-session, Recommendations). No Excel formulas (all values are displayed data); no native charts (the coverage matrix is itself the heatmap). Brand tokens apply to fill colors, border colors and state colors only — XLSX uses the reader's system fonts, not embedded fonts.
-
Filename convention: <slug> = descriptive domain/scope (as defined in step 1). Internal files (quality-report.md) stay without prefix.
-
Verify each file exists on disk with ls -lh output/<folder>/ after each writer skill returns. If a file is missing, regenerate — do NOT report partial success to the user.
6. Post-generation verification
For file formats:
ls -lh output/<folder>/ — confirm every expected file is present with non-zero size.- If any file is missing or zero-size: regenerate that format before returning to the user.
7. Final message to the user
After generation, present in chat:
- Generation confirmation (for Chat: the report itself; for files: folder path + list of generated artefacts with sizes).
- 2-3 key points from the report (e.g. top coverage gap, most critical KO rule, biggest improvement vs a previous report if available).
- Follow-up question (create rules for the detected gaps, expand scope to other tables, schedule periodic re-evaluation, etc.).
