Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› hri-writing-style
hri-writing-style
GitHub指导撰写或修改ACM/IEEE HRI会议论文，面向跨学科读者。强调以人机交互为核心，明确研究问题、假设与测量，清晰描述机器人具身形态，确保论点与研究规模匹配，并严格遵循8页篇幅限制。
Trigger Scenarios
撰写HRI会议论文引言
修改论文结构以突出人机贡献
澄清实验设计与假设
精简论文以满足页数限制
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill hri-writing-style -g -y
SKILL.md
Frontmatter
{
"name": "hri-writing-style",
"description": "Use when drafting or revising an ACM\/IEEE HRI full paper's structure and prose for an interdisciplinary audience — leading with the human-robot contribution and research questions, making hypotheses and measures explicit, describing the robot embodiment, keeping claims proportional to the study within the 8-page acmart budget, and writing so the chosen track's contribution type is unmistakable."
}
HRI Writing Style
HRI reviewers come from robotics, HCI, psychology, and design at once. A sentence that a roboticist
finds obvious may be opaque to a psychologist and vice versa, so HRI prose must be legible across
disciplines while making a rigorous contribution its home track can judge. You have 8 pages
(including figures, excluding references) in the ACM acmart template — a tight budget that rewards
a clear arc and punishes throat-clearing. This skill builds the paper for that audience and budget.
The HRI first-page arc
Lead with the interaction, not the machinery:
- The interaction problem — a question about what happens between a person and a robot, in one or two sentences, in the first breath.
- Why we do not yet know — the gap: what prior work measured, and why it does not answer this question (often: stated attitudes measured where behavior matters, or systems evaluated without people).
- The contribution, stated for your track — a designed behavior/system/interface, a finding about people, a method, or a concept — named as the kind of contribution your track reviews.
- The study — an embodied robot, real participants, a design a reviewer can judge, evidence proportional to the claim.
- What it means for designing interactive robots — the implication, concrete.
See resources/worked-examples/01-introduction.md for a fictional before→after that executes this
arc.
Make the study legible: state design, hypotheses, and measures
HRI's empirical reviewers expect the scaffolding to be explicit, ideally by the end of the introduction or the top of the method:
- Design — between-subjects, within-subjects, or mixed; conditions and what is manipulated.
- Hypotheses or research questions — stated as predictions (H1, H2...) for confirmatory work, or as RQs for exploratory/qualitative work. Do not disguise an exploratory study as confirmatory.
- Measures — which construct each dependent variable operationalizes, and whether the instrument is validated (e.g., a named trust/anthropomorphism scale) or bespoke (justify it).
- Robot and task — enough that a reader pictures the embodiment and the interaction (and point to the video).
A paper that hides its design until Section 4 forces reviewers to reverse-engineer it, which reads as evasion.
Describe the robot embodiment (and lean on the video)
HRI is about embodied interaction, so the robot is not a detail:
- Name the platform and the relevant capabilities/behaviors; show it in a figure and the video.
- Distinguish what was autonomous from what was teleoperated / Wizard-of-Oz — reviewers will
look for this, and hiding it is a credibility risk (see
hri-experiments). - Describe the interaction as it unfolds in time; static prose under-serves a temporal phenomenon,
which is exactly what the video is for (see
hri-supplementary).
Claims proportional to the study
The fastest way to lose an HRI reviewer is a claim the study cannot support:
- A single lab study with one robot and short exposure supports a bounded claim, not a universal law about "robots." Say "in this task, with this robot," and frame generality as future work.
- Report effect sizes and confidence intervals, not just p-values; "significant" without a magnitude is uninformative about whether the effect matters for design.
- For qualitative work, claim insight and transferability, not frequency; report your analysis method (e.g., thematic analysis) and reflexivity, not a fake sample size.
- Match the verb to the evidence: "suggests," "indicates," "in our study" — not "proves."
Write for the interdisciplinary reader
- Define discipline-specific terms once. A roboticist's "policy," a psychologist's "valence," and a designer's "probe" each need a gloss for the other two-thirds of the room.
- Justify constructs, don't assume them. Explain why the measured construct (trust, warmth, competence, mind perception) is the right one for the interaction you study.
- Prefer plain figures. A clear condition diagram and a results plot with error bars communicate across fields better than dense tables.
Budget discipline (8 pages, acmart)
- The template is fixed; recover space by cutting, not by shrinking margins or fonts (a
formatting violation is a desk-reject risk — see
hri-submission). - Move detailed instruments, full model tables, and extended qualitative excerpts to the
supplement, keeping in the body only what a reviewer must read to judge the paper (see
hri-supplementary). - References do not count against the 8 pages — but the body does, including figures. Budget figure space deliberately.
Anti-patterns HRI reviewers flag
- System-first framing — leading with a capability and accuracy, burying the human question
(route-signal that this may be an ICRA/IROS paper; see
hri-topic-selection). - Scale dump — administering many questionnaires and reporting whatever turned out significant, with no pre-registration and no correction.
- Undisclosed Wizard-of-Oz — presenting teleoperated behavior as autonomous.
- Liking as outcome — using a vague "users liked it" in place of the behavioral or theory-linked outcome the claim needs.
- Over-signposting — a paragraph of "Section 2 does X" standing in for an argument.
Output format
[First-page arc] problem → gap → contribution(track) → embodied study → design implication: present?
[Study legibility] design / hypotheses-or-RQs / measures stated early? yes/no
[Embodiment] robot described + video referenced? autonomy vs WoZ clear?
[Claim proportionality] claims bounded to the study? effect sizes reported?
[Budget] pages used (body incl. figures / refs) · formatting clean?
[Revisions] <ordered edits with section pointers>
Version History
- 9f86f09 Current 2026-07-19 15:51
