SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsacademic-topic-research-agent

academic-topic-research-agent

Writing & research

View repository

Updated 10-16 12:54

About this skill

Academic Topic Research Agent

Chinese quick start

  • How to invoke: Just tell the assistant your research idea (even one sentence), and add trigger phrases such as "make a research report", "write a v2-style pre-research", "help me make the topic solid".
  • Minimum information to provide: research topic/core question; target conference or journal; depth preference (quick/standard/deep). Other information (year, data type, must-include/must-avoid keywords, time budget, existing PDFs) the assistant will proactively ask about.
  • What you will get: a v2-style Chinese research report (Markdown + LaTeX, with PDF attached when TeX is available locally), 3-8 title candidates, 1-3 abstract candidates, a literature metadata matrix, a data/research feasibility judgment (Pass / Weak Pass / Fail), and a next-step execution plan.
  • Example trigger phrases：帮我把"用 AI 辅助高中物理老师做课堂决策"这个想法做成 CHI 风格的中文 v2 调研报告，standard 深度。
  • Want researcher-grade (CHI/CSCW level): add one more keyword, for example researcher-grade / 深度打磨 / 构念压力测试 / CHI 级 / CSCW 转身 / 期刊投稿级。deep mode is researcher-grade by default.

Researcher-Grade Mode

Normal mode produces an "OK research plan". Researcher-grade mode produces a "plan that can go directly into CHI/CSCW-level discussion" — after Modules A–E (research + candidate construction), it runs a segment of Module H: construct stress testing + 8-angle critique + rewrite loop + venue fit decision, pushing the candidate writing stage from "filling in TODO tables" to "research-grade constructs".

Depth Behavior
deep Enabled by default. Module H automatically runs 2 rounds of critique → rewrite. All 14 elements are done.
standard Off by default. Enabled when researcher-grade / 深度打磨 / 构念压力测试 / CHI 级 / CSCW 转身 / 期刊投稿级 appears in the user message, Module H runs 1 round. 11 of the 14 elements are done.
quick Skip Module H, but still enforce 1 Neighbor Concept Differentiation Table(max 3 lines), written to drafts/contribution_candidates.md。

Module H detects 8 types of structural failure: headline stealing the show, whether the construct is a neighbor concept renamed, whether the ladder secretly compresses multiple dimensions, whether the formative study is performance or inquiry, whether the ground-truth is stable, whether deployment cost is hidden as background, whether the naming pre-embeds the answer, whether the venue is split-personality. Each angle comes with trigger / fail / fix, and early stops when ≤ 1 angle fails to avoid fake critique theater. The full protocol is in references/construct_critique.md; the 14-element checklist is in references/researcher_grade_checklist.md; the venue decision is in references/venue_fit.md(covering CHI / CSCW / UIST / LAK / L@S / AIED / EDM / IJHCI / IJCAI / NeurIPS + representative Chinese journals).

Core Contract

Output language: default Chinese (default: Chinese). All final reports, drafts, README paragraphs, and answers to the user are written in Chinese. Use English only in the following scenarios: paper / conference / journal names, technical terms, code identifiers, and English versions explicitly requested by the user. This rule is enforced consistently on Claude Code and OpenAI Codex — a different runtime does not change the default language.

Use this skill as a research development kit, not a closed or rigid workflow. The skill provides a reusable report skeleton, evidence standards, search depth presets, and optional modules. The assistant should choose and combine modules according to the user's goal, source material, time budget, and available tools.

Never fabricate papers, datasets, citations, participant counts, metrics, results, PDF files, or study findings. Record uncertain material as Unknown, TODO, or Requires Manual Verification.

Platform Targets

This skill runs on multiple agent platforms. Both Claude Code and OpenAI Codex read SKILL.md (this file) plus the references/ and assets/ it points to. The split between platforms is only in auxiliary metadata and install paths:

  • agents/openai.yaml is Codex-only auxiliary metadata: UI fields (display_name, short_description, default_prompt), invocation policy (allow_implicit_invocation), and tool dependencies. Codex reads this for the app UI and the implicit-invocation matcher; it does not replace SKILL.md.
  • Install path: Claude Code uses .claude/skills/; Codex uses .agents/skills/. The same SKILL.md works in both — use $skill-installer, a symlink, or a copy to publish a Claude Code skill into Codex.
  • Interactive question UI: Claude Code uses the AskUserQuestion tool; Codex v0.119.0+ uses the ask_user_question tool. Both render card-style selectors with single/multi-select. See references/intake_questions.md Format B for the shared JSON schema.

When editing this skill, keep the name field and the core capability summary in sync between SKILL.md (here) and agents/openai.yaml. Shared assistant-facing docs in references/ and SKILL.md should stay platform-neutral and say "the assistant" rather than naming a specific runtime; platform-specific wording is OK inside agents/openai.yaml.

Intake Protocol

Before searching literature, downloading PDFs, or writing any report section, make sure you have enough input from the user. If the user's first message does not cover at least three of the following four core fields, pause and ask them in one consolidated turn — do not silently guess defaults and start searching.

Core intake fields:

  1. Research topic or core question — what the project is actually about.
  2. Target conference or journal — venue family (CHI, LAK, L@S, AIED, EDM, CSCW, UIST, IJHCI, IJCAI, …) or journal.
  3. Depth preference — quick / standard / deep (default standard when unclear).
  4. Data or research type — public benchmark, classroom data, interview/Wizard-of-Oz, annotation, simulated vignette, etc.

Strongly-recommended follow-up fields (ask in the same turn when natural):

  • Year range (default last 5-7 years for fast-moving venues)
  • Must-include keywords / must-avoid keywords
  • Is there already a proposal / .tex / PDF / notes
  • Time budget (decides whether to run deep mode)
  • Output language preference (default Chinese)

Use references/intake_questions.md for the exact question script. It provides a Markdown-numbered-list fallback (Format A) and a native question-card JSON version (Format B) that works on Claude Code (AskUserQuestion tool) and OpenAI Codex v0.119.0+ (ask_user_question tool). Prefer Format B whenever a native card tool is available; fall back to Format A only when no native question tool exists.

Codex single-turn handling: On Codex (and any single-turn-per-call runtime), the first turn ends after you post the intake questions via ask_user_question. On the next user reply, do not re-enter the Intake Protocol: parse the answers, write config/topic_input.yaml (and topic_lock.yaml when boundaries matter), summarize the understanding in 2-3 lines, then immediately start Module B (Metadata) in the same turn and drive all subsequent modules to completion without further pause unless a fatal blocker (missing critical data, sandbox denial) actually appears. Treat the user's intake answer as the explicit green light to run the full pipeline.

After the user answers, immediately write config/topic_input.yaml and (when boundaries matter) config/topic_lock.yaml. For quick tasks, embedding the lock as an in-report section is acceptable.

Depth Presets

Choose a depth before searching. If the user does not specify depth, default to standard.

  • quick: 10-20 candidate papers/systems/datasets. Use for rapid direction checks, title polishing, and short advisor memos.
  • standard: 30-50 candidate metadata records. Use for normal pre-research packages with literature map, dataset feasibility, title/abstract drafts, and v2-style report.
  • deep: 80-120 candidate metadata records. Use when the user asks for deep research, systematic preparation, "100 papers", "100 PDFs metadata", "complete preliminary research", or broad venue mapping.

For deep, collect metadata and OA PDF URLs first. Download only legally available open-access PDFs or user-provided PDFs. Do not treat downloading 100 PDF files as mandatory.

Quality threshold (not just count): a depth target is met only when at least ⅓ of records reach keep or background screening status. Fifty candidate-only rows do not satisfy standard.

Composable Modules

Use modules flexibly. A deep task usually uses most modules; a quick task may use only a few.

Intent and Boundary Module

  • Read user-provided title, proposal, .tex, notes, or PDF.
  • Run the Intake Protocol above when input is thin.
  • Preserve the user's original research intent.
  • Create config/topic_lock.yaml and/or a concise topic-lock section only when it helps prevent drift.
  • If later search/design clearly deviates from topic_lock.yaml, write reports/05_topic_drift_warning.md flagging the deviation before continuing.

Literature Metadata Module

  • Read references/search_protocol.md and respect config/sources.yaml (only use sources with enabled: true).
  • Build query groups and collect paper/system metadata at the selected depth.
  • Write or update data/processed/paper_matrix.csv or data/processed/metadata_candidates.csv.
  • Record key queries and evidence decisions in logs/evidence_trace.jsonl.
  • Optional: copy config/tag_schema.example.yaml to config/tag_schema.yaml and replace the example T-tags with topic-specific ones before bulk coding.

Dataset and Study Feasibility Module

  • Read references/dataset_protocol.md.
  • Search real public datasets, benchmarks, repositories, dataset papers, and study alternatives.
  • Write data/processed/dataset_candidates.csv for standard/deep tasks.
  • Classify feasibility as Pass, Weak Pass, or Fail when the user needs a decision.

Research Design Module

  • Turn evidence into research questions, contribution candidates, system/method scope, user study or experiment plan, risk table, and next steps.
  • Keep model/data claims separate from user-facing or venue-facing contribution claims.

Construct Stress Test and Critique Refinement Module (Module H)

  • Run when researcher-grade mode is on (deep automatic; standard with keyword trigger).
  • Read references/construct_critique.md (8 angles), references/researcher_grade_checklist.md (14 elements), references/venue_fit.md (full venue table).
  • Produce drafts/construct_stress_test.md, drafts/critique_round_<N>.md (or _passed.md), drafts/construct_rewrite_<N>.md, drafts/venue_fit_decision.md.
  • 2 passes for deep, 1 pass for standard (when triggered); skipped for quick.
  • Pass condition: if ≤ 1 angle fails, early-stop and write _passed.md rather than fake-rewriting.
  • If the rewrite changes scope, update config/topic_lock.yaml and append to reports/05_topic_drift_warning.md.

Title and Abstract Module

  • Generate 3-8 title candidates and 1-3 abstract variants when useful.
  • Bind strong abstract claims to evidence or clearly frame them as proposed work.
  • In researcher-grade mode, use the spine chosen in construct_rewrite_<latest>.md as the basis for every title candidate; fill the Headline Competition Audit table in drafts/title_candidates.md; fill the Construct/Differentiation/Quantified-Placeholder sentence fields in drafts/abstract_candidates.md.

v2-Style Report Module

  • Read references/report_style.md.
  • Use assets/report-blueprint/ as the minimal writing skeleton.
  • Produce reports/final_topic_report.md and reports/final_topic_report.tex.
  • Compile reports/final_topic_report.pdf only when a local TeX toolchain is available. See references/report_style.md "Environment Notes" for required components (xelatex + ctexart + CJK fonts; Pandoc optional).
  • When a verdict memo is needed in addition to the full report, write reports/06_final_recommendation.md — a short advisor-facing decision page ("advance / narrow down / redo" + 3-5 reasons).
  • In researcher-grade mode, also integrate the 13 structural blocks from assets/report-blueprint/v2_researcher_grade_extensions.md (and .tex for LaTeX) using the integration map at the top of that file. Required elements per depth tier are listed in references/researcher_grade_checklist.md; missing required elements should be written as visible TODO lines.

Final Output Standard

For a full package, produce:

  • v2-style Chinese research-plan report in Markdown and LaTeX.
  • PDF only when local TeX tooling exists.
  • title polishing options.
  • abstract candidate(s).
  • metadata matrix or evidence table appropriate to the selected depth.
  • dataset/study feasibility judgment.
  • next-step plan and unsafe-claim list.

The final report should feel like 网页版个人调研/v2.tex: strong core judgment, concrete examples, prior-work positioning, data feasibility, research design, risks, and a practical timeline. The section skeleton is a starting point, not a cage. Merge, rename, or omit sections when the topic demands it, as long as the final report still explains the research direction clearly.

Resource Map

  • references/workflow.md: modular execution guide (includes Module H for researcher-grade mode).
  • references/intake_questions.md: dual-format question script for the Intake Protocol.
  • references/search_protocol.md: quick/standard/deep metadata and legal PDF protocol.
  • references/dataset_protocol.md: dataset and study feasibility checks.
  • references/report_style.md: v2-style Chinese report structure, writing preferences, and environment notes.
  • references/schemas.md: configs, metadata matrices, evidence logs, and report artifacts.
  • references/evidence_rules.md: claim safety and evidence labels.
  • references/construct_critique.md: 8-angle critique protocol for Module H, with trigger/fail/fix per angle and early-stop pass condition.
  • references/researcher_grade_checklist.md: 14 structural elements every researcher-grade final report must carry; tiered by depth.
  • references/venue_fit.md: full venue decision tables for CHI / CSCW / UIST / LAK / L@S / AIED / EDM / IJHCI / IJCAI / NeurIPS plus Chinese journals; templates for drafts/venue_fit_decision.md.
  • assets/report-blueprint/: copyable v2-style minimal Markdown/LaTeX report skeleton.
  • assets/report-blueprint/v2_researcher_grade_extensions.md and .tex: 13 additive blocks appended to the v2 report when researcher-grade mode is on (opening thesis box, headline audit, reverse pitch, 3D operationalization, archetype table, neighbor differentiation, venue strategy, RQ matrix, alternative outcomes, deployment cost, reviewer attack, demand audit, elevator pitch); LaTeX includes tcolorbox macros (thesisbox/bluebox/redbox/greenbox).
  • assets/project-template/: starter project structure.
  • assets/project-template/config/sources.yaml: per-source on/off list — respect this when searching.
  • assets/project-template/config/tag_schema.example.yaml: illustrative T-tag set; copy to tag_schema.yaml and customize before use.
  • assets/project-template/drafts/construct_stress_test.md, critique_round_{1,2}.md, construct_rewrite_{1,2}.md, venue_fit_decision.md: Module H draft templates.
  • assets/project-template/scripts/render_report.py: render Markdown to styled LaTeX/PDF.
  • assets/project-template/scripts/workflow_checklist.py: inspect recommended artifacts by module (python ... --list).