dsh-academy

When drafting literature reviews or polishing paper footnotes with an LLM, the thing most likely to get you rejected at peer review is rarely the idea — it is the behaviour: a fabricated DOI, model inference dressed up as a literature claim, or a journal title mangled in a Chinese draft. dsh-academy targets exactly that layer. It registers no tools and no skills; its only job is to inject an academic-behaviour contract into the system prompt so the agent runs a researcher-level self-check before putting word to paper.

Three areas are covered. The research persona positions the reader as someone who needs a manuscript that survives peer review, not a paragraph that merely sounds right. Academic-integrity rules forbid inventing papers, authors, years, journals, DOIs, citation counts, or data; require separating what came from the literature from what is the model's inference; demand that every citation be verifiable; and leave judgement calls to the researcher. Language conventions call for answering in the user's language, keeping titles, journal names, and terminology in their original form, and following Chinese academic phrasing conventions when writing in Chinese.

It is aimed at researchers working on literature reviews, manuscript drafts, or academic presentations. On its own it shifts the agent's scholarly tone immediately; paired with dsh-ai4scholar (38 literature-search tools) or dsh-research (peer-review, introduction, formatting, and reference-audit skills), the integrity rules become fully enforceable because the agent can actually look up papers.