research-architect-audit
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
用别的 agent?下载 .zip 解压,把文件夹放进它的技能目录
~/.claude/skills/(项目级 .claude/skills/)~/.codex/skills/npx oh-my-skill add mmthebest/research-architect/research-architect-auditcurl -fsSL https://oh-my-skill.com/install.sh | bash -s -- mmthebest/research-architect/research-architect-auditnpx oh-my-skill verify mmthebest/research-architect/research-architect-audit怎么用
技能原文 SKILL.md
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-literatureorresearch-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.