| name | academic-style |
| description | Use when drafting or editing academic prose — topic sentences, paragraph structure, APA-style citations, active voice, tense, hedging conventions, term definitions. Fires eagerly on any .qmd / .Rmd / .md / .tex file in an academic-research project. Do NOT use for the revision workflow — use `manuscript-revision` + `/critic-loop`. Do NOT use for factual or citation accuracy — use `grounded-citations` / `empirical-integrity` / `fact-check`. |
Academic style
No pre-flight, no bootstrap by design. This is a doctrine
skill — pure prose conventions. It does not call MCPs, does not
install project scaffolds, and does not need /setup to have run.
The procedural skills (zotero-operations, systematic-review,
fact-check, critic-loop) gate themselves on
check_configured.py; this one stays out of the way.
Core rule
Academic prose follows a small set of conventions that the agent applies
during drafting, not only at revision time. These are the rules the
critic-loop argument critic checks later — but writing them in
correctly the first time reduces how much the critic loop has to
fix.
This skill governs style and conventions only. It does not govern:
- Citation sourcing (Zotero-backed, consulted, claim-supporting) →
grounded-citations rule-book.
- Quantitative claims (numbers from pipeline files) →
empirical-integrity rule-book.
- When and how to revise a draft →
manuscript-revision + the critic-loop skill.
Citations (APA-style)
Formatting only; sourcing is governed by grounded-citations — follow
both. Format conventions:
- Inline:
[@key] produces "(Smith, 2019)".
- Parenthetical with multiple:
[@key1; @key2] produces
"(Jones, 2020; Smith, 2019)".
- Narrative:
@key [-@key] produces "Smith (2019)" / "Smith's (2019)".
- Three or more authors: render as
et al. automatically via the
CSL; do not hand-type "et al.".
- Cite the most specific pinpointable source. When a claim could come
from a review or from the primary paper, cite the primary.
Voice and tense
- Active voice where it strengthens clarity: "We coded 1,243
papers" over "1,243 papers were coded". Passive is acceptable
when the actor is unknown or unimportant.
- Methods in past tense (this study's actions): "We ran a pilot
search against Scopus".
- Theory and established findings in present (ongoing truth):
"Self-efficacy predicts persistence".
- Discussion is present when interpreting ("Our findings
suggest") and past when describing what this study did ("We
found").
Hedging calibration
Strong claims require strong evidence. Match hedging to design:
- Cross-sectional data warrants "associated with", "is correlated
with", "differs across" — never "predicts", "causes", or "leads to".
- Observational longitudinal data without identification strategy
warrants "is longitudinally associated with" or "precedes" — still
not "causes".
- RCT / natural experiment / well-identified quasi-experiment
warrants causal language.
- Don't hedge findings the field considers well-established (this
signals false novelty). Don't over-claim findings that are novel.
(The critic-loop method critic flags violations at revision time —
writing to this standard up-front means the critic has less to catch.)
Structure
- Empirical papers: IMRAD (Introduction, Methods, Results,
Discussion). Each section scope-checked against Introduction's
stated research question.
- Review papers: synthesis over enumeration. The text must
analyse across cited studies, not march through them one at a
time. Long stretches of "Smith (2019) found X. Jones (2020) found
Y. Kim (2021) found Z." are a red flag — replace with thematic
prose that names patterns, tensions, or cumulative findings and
cites multiple papers per claim.
- Topic sentences open paragraphs. Signposting at section
transitions ("Having established X, we now turn to Y").
Terms and acronyms
- Define terms on first substantive use, not first mention.
- Expand acronyms on first occurrence: "better BibTeX (BBT)",
then
BBT thereafter.
- Consistent terminology throughout. Don't switch between "growth
aspirations", "growth intentions", and "growth motivation" for the
same construct without explaining the distinction.
Red flags
- You are writing "Smith (2019) found X. Jones (2020) found Y." chains
in a review paper — replace with thematic synthesis citing multiple
papers per claim.
- You are hedging a finding the field considers well-established —
drop the hedge.
- You are over-claiming from cross-sectional data ("predicts",
"causes") — switch to associational language.
- You introduced an acronym without expanding it.
- Terminology drift between sections — same construct, different
names.
- You are using passive voice where the actor matters and is known.
- You are adding an
et al. by hand instead of relying on CSL
rendering from the BBT key.