‹ 首页

research-architect-audit

@mmthebest · 收录于 1 周前

Audits reference-logic extraction, project adaptation, design-family fit, evidence traceability, citation integrity, inference boundaries, copying risk, and draft completeness.

适合你,如果需要系统检查研究草稿的引用和逻辑漏洞

/ 下载安装
research-architect-audit.skill双击,或拖进 Claude 桌面版 / Cowork,即完成安装↓ .skill↓ .zip
用别的 agent?下载 .zip 解压,把文件夹放进它的技能目录
Claude Code~/.claude/skills/(项目级 .claude/skills/)
Codex CLI~/.codex/skills/
Cursor自动读取上面两处目录
其他工具见其文档的「skills」目录;两个下载是同一份文件,只是名字不同
/ 通过 npx 安装 校验哈希
npx oh-my-skill add mmthebest/research-architect/research-architect-audit
/ 通过 bash 安装
curl -fsSL https://oh-my-skill.com/install.sh | bash -s -- mmthebest/research-architect/research-architect-audit
/ 已经装过?验证本机副本,不用重装
npx oh-my-skill verify mmthebest/research-architect/research-architect-audit
安装目标可用 --agent / --scope 或 --to 明确指定;省略时只会在唯一已存在的 agent 目录上自动选择,零命中或多命中会停止并提示。content_hash 缺失或不一致均拒装。
63GitHub stars
~623最小装载
~623含声明引用
~667文本包总量
镜像托管

怎么用

技能原文 SKILL.md作者撰写 · MIT · 4628ef0

Research Architect Audit

Use this branch before declaring the first draft complete.

Inputs

Read all available artifacts under:

paper_output/

Especially:

project_config.json
exemplar_logic_profile.md
exemplar_adaptation_plan.md
confirmed_research_spine.md
study_design_brief.md
study_component_registry.md
evidence_bank.md
claim_register.md
evidence_display_map.md
citation_support_bank.md
section_blueprints.md
writing_rationale_matrix.md
first_draft/main.md

Use compatible legacy artifacts when the canonical file is absent.

Required outputs

Create:

paper_output/audit_report.md
paper_output/revision_queue.md
Audit gates

| Gate | Blocking Failure Examples | |---|---| | Target-reference logic | target paper summarized without extracting load-bearing research moves or strengths | | Adaptation validity | topic nouns changed while design assumptions, context, or evidence needs remain unexamined | | Research spine clarity | no bounded phenomenon/tension/question/contribution/scope | | Feasibility | claim depends on evidence, access, participants, sources, or warrants that are unavailable and unplanned | | Literature grounding | sources are listed without mapping concepts, theories, debates, methods, evidence standards, or gaps | | Design-family fit | validity criteria or manuscript logic come from an incompatible research tradition | | Study-design validity | selection, evidence generation, analysis, comparison/warrant, credibility, or inference boundary is missing | | Evidence traceability | central claims lack evidence IDs, citation IDs, locators, or explicit limitation language | | Evidence-display logic | display has no argument function, hides context, or states a stronger claim than the evidence permits | | Citation integrity | citation is invented, unverifiable, irrelevant, or used beyond its support | | Non-copying | wording, source-specific claims, data, display content, or rhetorical sequence is too close to the exemplar | | Draft completeness | the design family requires methods, positionality, source criticism, integration, counterargument, limitations, or findings that are absent | | Inference-boundary control | causal, prevalence, transfer, theoretical, normative, mechanistic, or policy claims exceed the available warrant |

Audit report schema

| Gate | Status | Evidence Checked | Problem Found | Required Fix | Upstream Owner | Severity | |---|---|---|---|---|---|---|

Severity values:

  • BLOCKED: required before the draft can be used;
  • MAJOR: weakens the central research logic or claim;
  • MINOR: improves clarity, transparency, or presentation;
  • PASS: no issue found.
Revision queue schema

| Priority | Fix | Affected Section | Required Upstream Artifact | Owner/Next Action | Verification Criterion | |---|---|---|---|---|---|

Rules
  • Route structural failures to the artifact that owns the decision.
  • Weak target-paper analysis returns to research-architect-literature.
  • Superficial or infeasible adaptation returns to research-architect-literature or research-architect-brainstorm.
  • Weak design-family logic returns to research-architect-design.
  • Weak evidence or claims return to research-architect-evidence.
  • Weak citation support returns to research-architect-citation.
  • Prose revision follows upstream correction.
按 MIT 许可原样转载,未经改动 · 在 GitHub 查看 →

评论

登录即可评论;带「已验证安装」的,是发布者名下有本店的安装或持有记录。