URU Thesis Reviewer

Iterative feedback skill for URU theses. Each invocation produces a review session folder with numbered Markdown files the author reads, accepts/rejects, and applies to the source document. File templates in TEMPLATES.md; formal/linguistic check catalog in CHECKS.md; URU norms in NORMAS_URU_2020.md.

When to use

• User shares a .docx (or a converted .md/.txt) of a URU thesis or chapter and asks for review, feedback, suggestions, or revision.
• User asks to check semantics, grammar, clarity, structure, formal compliance, citations, or bibliography quality.
• User asks to compare against URU norms.

Do not use for:

• Editing the source document directly (this skill never touches the .docx).
• Evaluation / grading (this is improvement feedback, not a verdict).
• Non-URU theses (use a generic editor — URU norms are baked in).

1. Operating principles

• Read-only on source. Never edit, rewrite, or replace the .docx. All output is review files the author applies manually.
• Sequential & ordered. Output files follow the thesis reading order with numeric prefixes (01-, 02-, …).
• One session per invocation. Each review goes in a fresh YYYY-MM-DD/ folder. Multiple sessions in one day → suffix -a, -b, … (2026-05-21-a/).
• Diff format, not prose. Show the exact text to remove (-) and the proposed replacement (+).
• Severity-tagged. Every block carries one of: [blocking] (norm violation or factual error), [suggest] (real improvement), [nit] (taste/polish).
• Cite the rule. When the suggestion is driven by URU norms, link the section (e.g., NORMAS §V. CITAS). For language, cite RAE if relevant.
• Bibliography: suggest, don't fabricate. When flagging a weak source, propose a concrete replacement using find-docs / WebSearch. If no credible alternative is found, mark as [suggest] needs stronger source — none found in search.
• Spanish output. Feedback prose and rewrites are in Spanish. Tags, headers, and metadata stay English for grep-ability.

2. Inputs

User provides:

1. Source path — .docx (preferred) or already-converted .md/.txt.
2. Output base path — directory where YYYY-MM-DD/ session folder is created.
3. (Optional) Scope filter — e.g., "only Capítulo II", "skip bibliography", "focus on grammar only".

If the user only provides the .docx and no output path, ask before proceeding.

3. Ingestion (.docx → reviewable text)

Claude cannot Read .docx directly. Pick the first available path:

1. Anthropic official docx skill (preferred). Preserves structure (headings, lists, tables, footnotes, comments). Check the available-skills list at session start.
2. Pandoc fallback — pandoc "<source>.docx" -t markdown -o "<source>.review.md".
3. Manual. Ask the user to paste content or "Save as .txt" from Word.

The intermediate .review.md (from pandoc) is scratch — don't commit it, don't put it in the session folder.

After ingestion, identify section boundaries by matching headings against URU's expected structure (§6.1).

4. Output structure

Sessions live under <output-base>/YYYY-MM-DD/ with one numbered file per thesis section (01-INDEX.md, 02-RESUMEN.md, …, 11-REFERENCIAS.md, 12-ANEXOS.md). Full folder layout and naming rules in TEMPLATES § 5.

5. File templates

Three templates drive the output:

• 01-INDEX.md — executive summary, severity counts, file map, cross-cutting themes. Template in TEMPLATES § 1.
• NN-SECTION.md — per-section diffs grouped by severity. Template + diff shape in TEMPLATES § 2.
• 11-REFERENCIAS.md — bibliography-specific structure (missing/orphan sources, weak-source replacements, format fixes). Template in TEMPLATES § 3.

For substantive issues, the + side is a prompt for the author, not rewritten text — see TEMPLATES § 4.

6. Review dimensions (what to check, in order)

For each section, walk these dimensions in order. Stop reading prose once a [blocking] structural issue makes downstream feedback premature (note it in INDEX and move on).

Dimensions split into two groups:

• Substantive (§6.1–6.2 below): does the research itself hold up? This is the headline value of the review.
• Formal & linguistic (CHECKS.md): URU norms, prose quality, citation hygiene. Six checklists — formal aspects, redacción, grammar, semantics, citations, bibliography.

6.1 Structural compliance (URU §II)

• Are all required preliminary sections present? (portada, frontispicio, índices, resumen, abstract)
• Are chapters in URU order? (I. Problema → II. Marco Teórico → III. Marco Metodológico → IV. Resultados → V. Propuesta opt.)
• Does each chapter have the expected sub-sections per URU template?
• Are conclusiones organized by objetivos específicos?

6.2 Substantive content (proposals, information, argument)

Evaluate the thesis on its own merits as a piece of research, independent of URU formatting. Read each chapter as a domain reviewer would.

Cap. I — El Problema

• Problem clearly defined and bounded? Can the reader state it back in one sentence?
• Justification specific and evidence-backed, or generic ("este tema es muy importante")?
• Objetivo general: singular, measurable, achievable within the thesis scope?
• Objetivos específicos: decompose the general into independently-verifiable steps? Actionable verbs (analizar, diseñar, evaluar), not vague (estudiar, conocer)?
• Delimitación: clear scope boundaries.

Cap. II — Marco Teórico

• Antecedentes: actually relevant to this thesis's problem, or padding? Each should connect explicitly.
• Bases teóricas: depth matches the thesis's needs (not textbook summary, not off-topic monograph).
• Currency: dominant sources reasonably recent (varies by field — flag pre-2010 in fast-moving areas).
• Coverage gaps: obvious schools, key authors, recent developments missing?
• Operational definitions: key terms defined precisely and consistently.

Cap. III — Marco Metodológico

• Tipo y nivel: chosen design (descriptiva / explicativa / correlacional) appropriate for the objectives? Mismatches are [blocking].
• Diseño: experimental / no-experimental / documental — justified or just declared?
• Población y muestra: defined, sampling method explained, size justified (statistically or by saturation).
• Instrumentos: validated? Reliability reported? If self-built, validation process described?
• Procedimiento: replicable from the description alone.

Cap. IV — Resultados

• Do results actually address each objetivo específico? Cross-check explicitly.
• Claims proportional to the evidence? Flag overreach ("se demuestra que…" when data only suggests).
• Tables/figures show what the text claims? Internal contradictions are [blocking].
• Statistical / analytical method appropriate for the data type.
• Discussion: results contextualized against the marco teórico.

Cap. V — Propuesta (if present)

• Derived from the results, not pre-decided.
• Feasibility addressed (resources, time, constraints).
• Validation plan: how would adoption be measured?

Conclusiones

• One conclusion per objetivo específico (URU expects this alignment).
• Each conclusion supported by results — no new claims introduced.
• No overreach beyond the studied population/scope.

Recomendaciones

• Actionable, not platitudes ("se recomienda seguir investigando").
• Addressed to identifiable audiences.

Cross-chapter coherence (highest-value check)

• The chain problema → objetivos → marco → metodología → resultados → conclusiones must be intact.
• Verify: every objetivo específico has