Content & Media Creation · Imbad0202/academic-research-skills
academic-research-skills
Orchestrator for the full academic research pipeline: research -> write -> integrity check -> review -> revise -> re-review -> re-revise -> final integrity check -> finalize. Coordinates deep-research, academic-paper, and academic-paper-reviewer into a seamless 10-stage workflow with mandatory, coverage-bounded integrity checks, two-stage peer review, and auditable quality-assurance artifacts. Triggers on: academic pipeline, research to paper, full paper workflow, paper pipeline, end-to-end paper, research-to-publication, complete paper workflow, 연구부터 논문까지, 연구 주제 설정부터 논문 완성까지, 논문 전체 워크플로, flujo de trabajo académico, investigación a artículo, flujo completo de artículo, pipeline de investigación completa, publicación de investigación, flujo de trabajo completo del artículo.
L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandNetwork AccessWrite FileDual-pass Audit
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.
1How It Works & Mechanism
It is not a single skill, but a 4-skill suite (academic-paper writing / academic-paper-reviewer peer review / deep-research study / academic-pipeline orchestration). All modes are consolidated into a single registry table: 27 modes, acting as a single source of truth; any mode changes must update this registry first.
MODE_REGISTRY.md
Single source of truth for all modes across the ARS suite. **27 modes** across 4 skills.
Counts are strictly aligned across plugin.json ('4 skills, 27 modes, 39 prompt roles') and the 8/11/6 breakdown table in docs/ARCHITECTURE.md §9 (verified item-by-item in this check).
The orchestration layer is intentionally designed to 'do no heavy lifting': it only performs stage detection, mode recommendations, dispatching, and state tracking, delegating substantial tasks entirely to the other three skills.
academic-pipeline/SKILL.md
It does not perform substantive work — it only detects stages, recommends modes, dispatches skills, manages transitions, and tracks state.
This self-imposed constraint is the prerequisite for enabling its 10-stage state machine (including mid-pipeline entry: jumping directly to Stage 4 if reviewer comments are already available).
The writing side decomposes the 8-stage pipeline into 12 role prompts, using a v3.6.6 generator-evaluator 4-call contract in full mode for physical isolation—splitting the same logic into four separate model invocations to prevent 'inspecting the manuscript first and then rationalizing the standards.'
academic-paper/SKILL.md
For each `academic-paper full` invocation, Phase 4 + Phase 6 expand from two single calls into four separate model calls.
Supporting discipline: the system prompt carries invariant policy text only, while dynamic artifacts are routed via user-content data delimiters ('- **System prompt carries invariant policy text only**').
Integrity gates are mandatory and strictly block the pipeline: Stage 2.5 and Stage 4.5 must execute the 7-mode AI Research Failure Mode Checklist, blocking execution if SUSPECTED or INSUFFICIENT EVIDENCE on critical modes is triggered.
academic-pipeline/SKILL.md
If any of the 7 modes is `SUSPECTED`, or if Modes 1/3/5/6 are `INSUFFICIENT EVIDENCE`, the pipeline **blocks**
Paired with the rule that users can pause at any time but cannot skip checks: '⚠️ IRON RULE: User can pause or exit pipeline at any time (but cannot skip integrity checks)'.
The review side consists of a 5-member panel (Journal-Fit + 3 peer reviewers + a fixed Devil's Advocate) and is hard-constrained to be read-only—manuscript edits are prohibited.
academic-paper-reviewer/SKILL.md
Reviewers MUST NOT modify the submitted manuscript.
Any CRITICAL issue raised by the DA must have a visible adjudication in the editorial decision; an unresolved issue prevents silent acceptance ('a validated or genuinely unresolved one blocks silent Accept finalization').
Writing scope enforcement does not rely on self-discipline but is enforced at the hook layer: PreToolUse intercepts Write/Edit/MultiEdit/Bash and forwards them to a Python guard; if the guard fails, it falls back to pass-through and will never emit an allow.
hooks/run_guard.sh
# Canonical pass-through output: no permissionDecision => falls back to the normal permission # flow (NEVER emit "allow" — that would skip every other permission rule).
The guard also protects its own infrastructure: INFRA_PROTECTED_GLOBS includes hooks/hooks.json, hooks/*.sh, and .claude-plugin/plugin.json, with annotations explicitly stating that disabling it neuters the guard.
Citation verification uses a deterministic four-index existence gate (Crossref / OpenAlex / arXiv / Semantic Scholar) with results cached locally, explicitly distinguishing between a user's self-reported 'I have read this' and independent evidence.
docs/DATA_FLOWS.md
The deterministic citation-existence gate (#182) fires at the Stage 2.5 / 4.5 integrity gates or on standalone `verify_passport.py` / `verify_citation` calls, cache-through by default.
Exact wording for the manual reading signal: 'This is a user statement, not independent evidence that a person read or understood the source.' (commands/ars-mark-read.md)
The network surface is self-disclosed and enforced via CI synchronization: any script adding a direct network import or any shell command adding curl must be registered in DATA_FLOWS.md beforehand, or CI will fail.
docs/DATA_FLOWS.md
a script that gains a *direct* network import, or a shell script that gains a `curl`, fails CI until it has a row here.
The same document defines strict honest boundaries: '**Nothing here publishes.** No ARS component is designed to submit, post, or upload your work anywhere autonomously' and '**The Claude session itself is not on this map.**'
The only default automated outbound call is the version check upon session startup: it curls upstream plugin.json for version comparison with a 24h cache, a 3s timeout, an option to disable with a single flag, and anti-poisoning version string validation.
scripts/ars_update_check.sh
REMOTE_URL="${ARS_UPDATE_CHECK_REMOTE_URL:-https://raw.githubusercontent.com/Imbad0202/academic-research-skills/main/.claude-plugin/plugin.json}"
Kill switch: `if [[ "${ARS_UPDATE_CHECK:-}" == "0" ]]; then`; the announce script notes that compact/resume paths structurally never run it ('compact/resume structurally never run it (no network mid-session)').
2Core Capabilities
01End-to-end 10-stage orchestration (research -> write -> integrity check -> review -> revise -> re-review -> re-revise -> final check -> finalize), supporting mid-pipeline entry and state tracking
02Academic writing across 11 modes: full/plan/outline/revision/revision-coach/abstract/lit-review/format-convert/citation-check/disclosure/rebuttal-audit, supporting 6 paper types, 5 citation styles, and bilingual abstracts
03Multi-perspective peer review across 6 modes (full/re-review/quick/methodology-focus/guided/calibration), with re-review following an 'evidence-before-persuasion' three-gate contract
04Deep research across 8 modes, including PRISMA systematic reviews with optional meta-analysis, risk-of-bias assessment (RoB 2 / ROBINS-I), and predatory journal screening
05Deterministic citation existence verification + 90-day local caching + stale-entry warning/full-rerun switch
06AI usage disclosure and journal policy adaptation (15 policy targets, including ICLR/NeurIPS/Nature/Science/ACL/EMNLP and ICMJE/NEJM/Lancet/JAMA)
07Socratic guided mode with a hard non-generative boundary: never auto-generates candidate research questions; exiting requires an explicit flag
08Auditable artifact chain: Material Passport, panel provenance replay verification, and process record PDFs
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| api | Semantic Scholar Graph API（引文存在性校验，key 可选，无 key 也工作） |
| api | OpenAlex API（引文存在性校验，支持 polite email） |
| api | Crossref REST API（DOI 元数据/存在性） |
| api | arXiv API（预印本存在性，无 key，≥3s 节流） |
| api | 跨模型盲评（opt-in，需 env + 逐会话同意；评审卡与论文元数据会发往外部 provider） |
| network | 会话启动版本检查（默认开启，curl 上游 plugin.json，3s 超时，ARS_UPDATE_CHECK=0 关闭） |
| cli | python3（守卫、校验、引文客户端等 ~431 个脚本；README 称核心技能是 prompt-driven，python 仅守卫与少数 opt-in 命令可选） |
| cli | Pandoc / tectonic + Source Han Serif TC（DOCX / APA 7.0 PDF 输出，可选） |
| package | 开发依赖（pytest / yaml / jsonschema），CI 内 pip install |
4Recommended Governance & Evidence
Suggested Controls for Safe Operation
For Agent Runtime
Graded/per-use authorization, sandbox recommended
For Administrators
Approval workflow, rollback support, full audit
Evidence & Triggers (File Paths & Excerpts)
- scripts/: Local script or command execution: hooks/run_guard.sh (triggered automatically via PreToolUse) -> scripts/ars_write_scope_guard.py: intercepts Write/Edit/MultiEdit/Bash
- 网络端点: Declared external network access: Default automated: SessionStart curls https://raw.githubusercontent.com/Imbad0202/academic-research-suite/...
- 本地工作目录: File I/O and artifact output: Hook-enforced writing scope: PreToolUse intercepts all Write/Edit/MultiEdit/Bash operations, evaluated by ars_write_scope_guard.py
Audit Notes: Classification rationale (based on 6 tiers: Orange if credential/env var reads or remote package dependencies are present): ① Genuine reads of multiple API credentials: S2_API_KEY, OPENALEX_API_KEY, CROSSREF_POLITE_EMAIL, OPENAI_API_KEY, GOOGLE_AI_API_KEY, ARS_OPENAI_COMPAT_API_KEY, OPENREVIEW_USERNAME/PASSWORD; ② Genuine outbound network calls: four academic index endpoints, optional cross-model providers (api.openai.com / generativelanguage.googleapis.com / custom base URLs), and default session-startup curl upstream update checks; ③ Hook-level automated execution and 431 Python scripts (running local repo code, not remote installs), writing to ~/.cache/ars cache with writing scopes governed by guards; ④ Supply chain surface: CI pulls dev dependencies from PyPI, optional Pandoc/tectonic. Mitigating factors (do not lower the tier, but must be disclosed): all four citation gate indices can be queried keyless (keys are optional); cross-model usage is opt-in and requires session-by-session consent; 'Nothing here publishes' and network flows are enforced by CI registration; writing scopes have hook guards and infrastructure self-protection; review workflows feature explicit untrusted-material clauses. Notes: Enabling this in Claude Code means SessionStart will access the network and PreToolUse will run local Python guards; pay attention to data exfiltration risks with cross-model options when processing unpublished manuscripts; licensed under CC-BY-NC-4.0, requiring separate licensing for commercial use.
5Second-Pass Verification
- [ok] skill.path（任务表标『待定位』）；套件条目指向哪个具体 skill — 本条目展示名『Academic Research Skills』对应的是 4-skill 套件，仓库根无集合级 SKILL.md。skill.path 现指向套件编排器 academic-pipeline —— 它是 research→write→review→revise→finalize 全流程的入口与状态机（10 阶段、强制完整性闸门、passport 续航），也是 plugin.json 描述的『Contract-audited academic research pipeline』最直接的承载者。其余 3 个并列 skill 为：academic-paper（12-agent 写作流水线）、academic-paper-reviewer（5 席评审面板）、deep-research（13-agent 研究流水线）；另注意仓库内 skills/ 目录只有 4 个指向上述目录的符号链接，是给按 skills/ 查找的宿主用的发现 shim。本 JSON 的 caps/assets 覆盖全套件，evidence 逐条指向各自子目录。
- [discrepancy] official_desc 取哪一份（套件无集合级 frontmatter description） — 套件在仓库根没有集合级 SKILL.md/frontmatter description，因此官方描述有两个候选口径：代表 skill academic-pipeline 的 frontmatter description（逐字原文，含 10-stage 编排与多语触发词），或 .claude-plugin/plugin.json 的集合级 description（'Contract-audited academic research pipeline for Claude Code: research → write → review → revise → finalize. 4 skills, 27 modes, 39 prompt roles…'）。本条目 official_desc 采用前者（与 skill.path 指向的具体 skill 一致，满足『official_desc = SKILL.md frontmatter 原文』的取证规则）；集合级口径已逐字记入 internal_assets 的 academic-pipeline/SKILL.md 条目说明，不丢失信息。注意该 description 本身在源文件里就是长触发词清单（中/韩/西多语），属原文形态而非我方截断。
- [discrepancy] 13-agent 声称 vs agents/ 实际文件数 — deep-research/SKILL.md 写 '## Agent Team (13 Agents)'，而 deep-research/agents/ 实有 14 个文件（多出 timeline_extraction_agent.md）。docs/ARCHITECTURE.md §9 把它归入 v3.9.4 时间线校验咨询层（'v3.9.4 temporal verification advisory layer (M1 timeline_extraction_agent + M2 5-pass verifier at Phase 4→5 + M3 IRON RULE + M6 first-party Crossref/pdftotext)'），即新增角色未回填到 SKILL.md 的 Agent Team 表——属文档滞后，不是能力夸大。其余三个 skill 的声称数与文件数一致（12/7/5）。
- [ok] 『39 prompt roles』是否有据 — 权威源 scripts/model_tiering_manifest.json 的 agents 数组恰 39 条（每条 path+tier：judgment/execution），shared/model_tiering.md 标题 '## Classification table (39 agents; frozen 2026-07-11, #517)'，且 scripts/check_model_tiering.py 把该 manifest 与磁盘 *_agent.md、文档表三方钉死。磁盘 42 个 *_agent.md 中未被 manifest 收录的 3 个正是根 agents/（插件可见，model: inherit），与 plugin.json 自述一致。
- [ok] 『14 workflows』与仓库实际 — scout 记为低风险未核（只见到 13 个名字）；本人 ls .github/workflows 得 14 个 yml（changelog-covers-merges / command-invariants / defer-label-gate / eval-harness / freshness-check / harness-retirement-monthly / platform-port-reminder / pr-closes-issue / pytest / release-cooldown / repository-hygiene / spec-consistency / tag-version-match / test-count-monotonic），与 ARCHITECTURE 的 'all fourteen workflows' 相符。
- [discrepancy] 引文闸门有效性（README 暗示 vs 自评） — 仓库自己的材料承认未被测量：docs/RISK_REGISTER.md 写 '**Evidence status**: `NOT_RUN` (capability matrix row `retrieval.citation_existence_gate`)' 与 '**Residual gap**: no measured hallucinated-citation catch rate'；examples/showcase/README.md 更自曝 'a manual WebSearch audit of all 68 references revealed 21 issues (31%) that survived three rounds of automated integrity checking'。因此本 JSON 只陈述机制存在，不背书其检出率——这属『实现真实、效果未证』。
- [discrepancy] 校准模式是否可用于线上评审 — academic-paper-reviewer/SKILL.md 自认 'a target-specific candidate measured profile labelled `application_status: NOT_WIRED_TO_LIVE_REVIEW`'，且线上评审与 Schema 6 包声明 NOT_CALIBRATED。若 README/宣传让人以为面板已校准，应以 SKILL.md 为准。
- [ok] 反注入条款的覆盖面 — 复核确认：仅 academic-paper-reviewer 及 shared/ground_truth_isolation_pattern.md 有显式条款（后者自认 'This is a guidance principle, not a runtime gate.'）；academic-paper / deep-research / academic-pipeline 三目录 grep injection|untrusted 零命中。已在 security.injection_surface 中按这一实况书写，未把评审侧的防线推广成整套件的能力。
6Verdict & Scenarios
Best for: Ideal for: researchers and graduate students who need to produce submission-ready manuscripts (including bilingual Chinese-English writing and submission disclosure workflows); research teams aiming to reproduce systematic reviews (PRISMA + RoB); and institutional users treating 'AI output reliability' as a core concern and willing to accept workflow overhead for gates and audit trails. Also serves as an engineering benchmark for 'building long workflows as a contracted and evaluated skill suite.' Best experienced in Claude Code (with hook-enforced writing scopes and 16 slash commands).
Not for: Not recommended for: ① Lightweight users merely wanting quick copyediting or a single blog chapter (excessive overhead); ② Air-gapped/internal network environments that strictly prohibit any default outbound network requests; ③ Commercial closed-source redistribution (licensed under CC-BY-NC-4.0); ④ Host environments lacking Node/Claude Code where hook-level write scope enforcement is required (falls back to prompt-level under Pi); ⑤ Users expecting 'one-click reliable papers'—the repository's own showcases prove that automated integrity checks can miss issues, making human review indispensable.
Install Agent-Ready
① Verified Mirror 2026-09-15
Option A · Direct DownloadDownload academic-research-skills.tar.gz
sha256:
Option B · Copy Guide for Agentc34ec0b645ba9a32…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit 7e019f5b25.