| name | academic-research-suite |
| description | Academic Research Skills adapted for Codex from Imbad0202/academic-research-skills. Use when the user asks for academic research planning, literature review, systematic review or PRISMA work, research question refinement, manuscript outlining, paper drafting, abstract writing, revision, citation checking, citation format conversion, AI disclosure, peer review, response-to-reviewer strategy, integrity checks, or an end-to-end research-to-publication pipeline. Also use for ARS slash-command style requests such as ars-plan, ars-lit-review, ars-reviewer, ars-full, ars-outline, ars-abstract, ars-revision, ars-revision-coach, ars-citation-check, ars-format-convert, or ars-disclosure. |
Academic Research Suite
Use this skill as a Codex-native entrypoint for the ARS source tree. ARS was written for Claude Code, so adapt its workflows to Codex instead of following Claude-only mechanics literally.
The canonical source should be available at references/source. If that directory is missing, run scripts/bootstrap-source.sh from the skill directory to clone https://github.com/Imbad0202/academic-research-skills.
Do not bulk-load the full source tree. Start from the routing below, then read only the relevant SKILL.md, agent files, references, templates, or scripts.
Quick Routing
Map user intent to one ARS family:
| User intent | ARS family | Start by reading |
|---|
| Explore a vague topic, refine RQ, collect evidence, fact-check, literature review, systematic review | deep-research | references/source/deep-research/SKILL.md |
| Plan, outline, draft, revise, polish, format, cite-check, abstract, AI disclosure | academic-paper | references/source/academic-paper/SKILL.md |
| Review an existing manuscript, methodology review, guided review, re-review after revision | academic-paper-reviewer | references/source/academic-paper-reviewer/SKILL.md |
| Full research-to-publication workflow or mixed artifacts across phases | academic-pipeline | references/source/academic-pipeline/SKILL.md |
Treat these ARS command names as aliases:
| Alias | Codex action |
|---|
ars-plan | academic-paper plan mode |
ars-lit-review | deep-research lit-review mode |
ars-reviewer | academic-paper-reviewer full mode |
ars-full | academic-pipeline full workflow unless the request is clearly single-phase |
ars-outline | academic-paper outline-only mode |
ars-abstract | academic-paper abstract-only mode |
ars-revision | academic-paper revision mode |
ars-revision-coach | academic-paper revision-coach mode |
ars-citation-check | academic-paper citation-check mode |
ars-format-convert | academic-paper format-convert mode |
ars-disclosure | academic-paper disclosure mode |
Codex Adaptation Rules
- Preserve ARS's quality stance: do not invent citations, data, results, reviewer findings, or methodological details.
- Replace Claude Code
Task/subagent assumptions with Codex execution. If multi-agent tools are available and the work is large enough, use them for independent passes. Otherwise run the roles sequentially in the current thread and label the role boundaries in your notes.
- Use Codex tools and skills normally. For current literature, policies, journal instructions, metadata, prices, laws, software versions, or anything likely to have changed, browse or use primary sources as required by the active system instructions.
- For citations, prefer DOI/publisher/official index metadata. If a source cannot be verified, label it as unverified instead of silently using it.
- When the user writes in Chinese, respond in Chinese by default. Translate search concepts into English when that improves scholarly retrieval.
- If the task is specifically Nature/CNS manuscript polishing, Nature/CNS citation insertion, data availability, reviewer response, figure preparation, or full-paper reading, consider combining this skill with the existing
nature-* Codex skills.
- If the user supplies cross-phase materials without a clear goal, clarify the target workflow before doing a single-phase transformation.
- Keep outputs useful rather than theatrical: concrete plan, evidence table, manuscript text, review report, revision matrix, citation audit, or pipeline status.
Source Navigation
Use rg against references/source to find the relevant protocol before opening large files:
rg -n "mode|Trigger|PRISMA|citation|review|revision|disclosure|pipeline|passport" references/source
Useful starting files:
| Need | Files |
|---|
| Mode selection and aliases | references/mode-map.md, references/source/MODE_REGISTRY.md, references/source/commands/ |
| Deep research details | references/source/deep-research/SKILL.md, selected files in deep-research/agents/ and deep-research/references/ |
| Manuscript writing details | references/source/academic-paper/SKILL.md, selected files in academic-paper/agents/ and academic-paper/templates/ |
| Peer review details | references/source/academic-paper-reviewer/SKILL.md, selected files in academic-paper-reviewer/agents/ and academic-paper-reviewer/references/ |
| Full pipeline and Material Passport | references/source/academic-pipeline/SKILL.md, academic-pipeline/references/pipeline_state_machine.md, shared/handoff_schemas.md |
| Deterministic checks and adapters | scripts/ars-source-scripts/, especially schema/lint/check scripts |
Operating Workflow
- Classify the request into one ARS family and mode.
- Read the corresponding source
SKILL.md.
- Read only the agent/reference files needed for that mode.
- State the intended workflow briefly if the job is substantial.
- Execute the work with verification appropriate to risk:
- research: source search, inclusion/exclusion notes, citation support grading
- writing: claim/evidence alignment and transparent assumptions
- review: severity-ranked findings with evidence from the manuscript
- pipeline: stage status, gates, unresolved decisions, next artifact
- Use scripts from
scripts/ars-source-scripts/ when deterministic validation is useful.
Guardrails
- Do not present ARS as an autonomous paper-writing service. Keep the human researcher responsible for research question, method, interpretation, and final claims.
- Do not convert a vague topic directly into a polished paper without first establishing scope, evidence, and assumptions.
- Do not use fabricated or memory-only references.
- Do not hide AI involvement in disclosure tasks. Draft honest, venue-appropriate disclosure language.
- Do not run destructive file operations on user manuscripts or datasets unless explicitly requested.
