| name | uru-thesis-reviewer |
| version | 0.2.0 |
| description | Continuous-feedback reviewer for URU (Universidad Rafael Urdaneta) theses. Emits ordered `.md` diff files (- old / + new) the author applies manually — never edits the `.docx`. Covers substance, structure, prose, citations vs `NORMAS_URU_2020`. Tracks issues across sessions via stable IDs and carry-forward. Use when the user shares a thesis or chapter. |
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:
- Source path —
.docx (preferred) or already-converted .md/.txt.
- Output base path — directory where
YYYY-MM-DD/ session folder is created.
- (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:
- Anthropic official
docx skill (preferred). Preserves structure (headings, lists, tables, footnotes, comments). Check the available-skills list at session start.
- Pandoc fallback —
pandoc "<source>.docx" -t markdown -o "<source>.review.md".
- 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 & cross-session tracking
4.1 Session folder
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), kebab-case (lowercase, dash-separated, no accents). Full folder layout and naming rules in TEMPLATES § 5.
01-index.md is mandatory every session, even one that only drafted prose and touched no tracked issue — a 5-line index (executive summary + "sin cambios en seguimiento de issues") beats a missing one. There is no separate cross-session registry file: 01-index.md is the record, session over session.
Any file the session produces that isn't a numbered section review — free-form drafting (redaccion-*.md), a handoff note — still gets listed in 01-index.md's Mapa de archivos, tagged by type (see TEMPLATES § 1).
4.2 Cross-session issue tracking
Each 01-index.md carries a Seguimiento de issues table (template in TEMPLATES § 1) that is the running state of every issue ever raised, not just this session's.
- Stable IDs. Assign each issue an ID at first discovery:
yyyy-mm-dd-<chapter>-<slug> (e.g. 2026-06-06-cap-ii-mcp-como-agente). On recurrence, reference the existing ID — never restate the finding as new.
- Status is orthogonal to severity. Severity (
[blocking]/[suggest]/[nit]) says how bad it is; status says where it stands: abierto (found, not addressed) → resuelto-en-borrador (fix drafted in a review file, not yet in the .docx) → aplicado (author confirmed it's in the .docx) → rechazado (author declined) → needs-decision (open question for the author/tutor). Only the author can move an issue to aplicado — Claude cannot verify the .docx was edited.
- Carry-forward, mandatory. At the start of every session after the first, read the previous session's
01-index.md Seguimiento table before writing anything new. Copy every issue not at aplicado/rechazado forward into this session's table unchanged unless its status actually moved. This is what makes the per-session file sufficient as a registry — skipping it is how issues silently go stale.
- Reversed decisions need a reason. If this session overturns or drops a prior decision (not just a status update — an actual reversal), add a one-line
**Por qué se revirtió:** next to that issue's row. A diff showing the old text gone is not enough.
- Tutor comment numbers drift.
.docx comment numbers renumber across edits. Always pair a referenced comment number with the verbatim quoted text it points to — never cite the number alone.
5. File templates
Three templates drive the output:
01-index.md — executive summary, severity counts, file map, cross-cutting themes, and the cross-session Seguimiento de issues table (§4.2). 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 a corresponding methodological step, a results subsection, and a conclusion.
- Verify: every conclusion traces back to a result, which traces back to a method, which addresses an objective.
6.3 Formal & linguistic dimensions
Formal aspects (URU §IV), redacción (URU §III), grammar/RAE, semantics & clarity, citations (URU §V), bibliography (URU §VI) — full per-dimension checklists in CHECKS.md.
7. Workflow
- Carry forward. If a prior session exists, read its
01-index.md Seguimiento de issues table (§4.2) before doing anything else. This seeds the current session's tracking table.
- Ingest.
.docx → reviewable text via §3. Identify section boundaries.
- Inventory. List every section found. Compare against URU expected structure (§6.1). Flag missing/extra sections as
[blocking].
- Scope. If the user specified a filter, restrict to those sections.
- Substantive pass first. Walk §6.1 then §6.2 across all chapters before touching prose. Substantive
[blocking] issues reshape downstream work — surface them early.
- Per-section formal/linguistic pass. Walk CHECKS.md §1–5 in order. Skip deep prose review on any section flagged with a substantive
[blocking] — note "prose review deferred pending substantive revision" in INDEX.
- Bibliography pass. Run CHECKS § 6 at the end — needs the full source list and may require web searches.
- Write files. Create
YYYY-MM-DD/ folder. Write section files in order. Write 01-index.md last, including the carried-forward + updated Seguimiento de issues table. The Resumen ejecutivo leads with substantive impression before formal observations. If the session hands off open work, write it as 00-handoff.md — always that name, never a variant.
- Report. Brief message to user: session path, severity totals, recommended starting point. Do NOT dump the diffs into chat — point to the files.
8. Reference
- URU norms (authoritative source): NORMAS_URU_2020.md in this skill's folder. Read it on first invocation of a session and keep it in working context.
- RAE for orthography & grammar.
- APA 7th where URU norms are silent (URU is APA-derived but Spanish-localized).
9. Out of scope
- Editing the
.docx directly. Author always applies changes.
- Rewriting entire sections from scratch — propose targeted diffs, not redrafts. If a section needs total rewrite, say so in
[blocking] and explain why.
- Plagiarism detection — flag suspicious unattributed phrasing as
[suggest] verificar fuente, don't claim plagiarism.
- Defense/jury preparation — separate skill.
- Translating Resumen ↔ Abstract — flag mismatches, don't translate.