nlr-workflow

Factual fidelity audit for academic literature review manuscripts. Cross-references every specific quantitative claim in the manuscript — metric values (AUC, accuracy, F1, BLEU, hazard ratio, etc.), dataset names, model/architecture names, and sample sizes — against the source extraction notes or paper summaries, then reports which claims are verified, which are mismatched, and which lack a traceable citation. Use this skill whenever a manuscript draft has been updated, before any style-polish pass, or any time you suspect numbers may have drifted from their source notes. Works with any markdown manuscript + extraction note workflow, regardless of domain or citation style. Invoke with /lit-audit.

Converts author-year citation placeholders in a manuscript draft to pandoc citekey format, matched against a BibTeX (.bib) file. Handles the full replacement pipeline: parse the .bib file to build an author+year → citekey index, scan every placeholder in the manuscript, auto-replace unambiguous matches, collect ambiguous cases and ask the user to resolve them as a batch, then save the cited manuscript. Use this skill whenever a manuscript contains [Author20XX] or [Author20XX_N] placeholders that need to be converted to [@citekey] for pandoc or Zotero workflow. Works for any markdown manuscript + BibTeX library combination. Invoke with /lit-cite.

Removes AI writing fingerprints from academic manuscript drafts. Scans the text for four categories of AI register problems — antithesis constructions ("not X but Y"), editorializing adjectives ("striking", "remarkable", "unprecedented", etc.), repetitive structural refrains, and systematic-review language in a narrative review (paper counts like "89 studies in this corpus", exhaustive-corpus assertions like "none of the 280 reviewed studies", and meta-sentences like "this review synthesizes 280 studies") — then rewrites only the excess instances to bring counts within target thresholds. Produces the revised manuscript plus a before/after change table. Use this skill after factual verification and before any final style polish, or any time a manuscript reads as AI-generated or as a systematic review when it should be a narrative review. Works on any markdown manuscript regardless of domain. Invoke with /lit-deregister.

Drafts a single named section of an academic literature review manuscript, guided by the word budget in the project outline, content from per-paper extraction notes, and strict quality constraints (citation density, numeric accuracy, register, em-dash policy). Use this whenever any section of a review manuscript needs to be written or re-written — introduction, body section, or conclusion. Invoke by naming the section: /lit-draft 3 (section number), /lit-draft 3.1 (subsection), or /lit-draft "Methods Integration" (section title). Automatically locates the project outline, reads relevant extraction files, verifies every cited number against the source notes, runs a register self-check, and saves the draft to the draft/ directory. Works for any literature review project that keeps a section outline and per-paper extraction notes, regardless of topic or journal target. Invoke with /lit-draft [section identifier].

Quantitative submission-readiness gate for academic manuscript drafts. Runs a battery of objective, tool-based checks — word count, em-dash frequency, citation density, abstract presence, section structure, figure callouts, unfilled placeholders, broken citation keys, and (for narrative reviews) systematic-review language patterns — then produces a structured pass/fail scorecard and a verdict. Unlike asking Claude to "check the manuscript", this skill uses Bash and Grep to measure the actual file, so the results are verifiable and not self-reported. Use before any submission-prep step, after all editing passes are complete, or any time you need a quick health check on a draft. Works on any markdown manuscript. Invoke with /lit-gate, or /lit-gate [word-limit] to set a custom word count target.

Session startup status check for academic literature review projects. Reads the project log, counts files in each workflow directory, infers the current phase, and prints a concise status panel — all in one shot. Replaces the manual "check the project before starting" habit that users frequently forget. Use this skill at the start of every session, when resuming after a break, when you're unsure where a project stands, or when a user says "let's continue the review" or "where were we?" without providing context. Works for any lit-review project structure. Invoke with /lit-status.