Revise existing academic drafts into clear, audience-facing paper prose while preserving evidence, technical meaning, uncertainty, and scope. Use when Codex needs to make a draft read like a paper rather than a lab report, memo, or internal note; convert project nicknames or internal nomenclature into reader-facing terminology; sharpen contribution framing without inventing novelty; or rewrite methods, results, limitations, and discussion prose for an academic audience. Do not use when the main task is paper planning, citation work, novelty review, or experiment design.
Run iterative adversarial review over research plans, experiment outputs, and drafts with claim ledgers, issue tracking, evidence checks, and explicit closure criteria. Use when asked to red-team a research artifact across multiple rounds, maintain issue state across revisions, or pressure-test whether revised results and prose actually support a claim. Prefer `research-paper-review` for an initial single-paper critique or OCR/extraction workflow, and prefer `research-rebuttal` when concrete external reviewer comments already exist and the task is to draft a venue response.
Run evidence-gated commercialization analysis for academic research by translating a bounded research asset into workflow pain, current workarounds, buyer and budget logic, commercialization paths, risk-retirement tests, and kill/pivot/continue decisions. Use when Codex needs to evaluate market pull for a paper, prototype, dataset, device, algorithm, lab result, or platform technology; distinguish user, buyer, operator, procurement, and budget roles; compare startup, licensing, partnership, service, or component routes; identify beachhead wedges; build validation sprints; or red-team unsupported commercialization claims.
Read a specified set of documents and distill them into a reusable, source-grounded memory pack. Use when Codex is asked to ingest project documents, papers, specs, notes, policies, or mixed corpora and create durable memory artifacts such as canonical facts, terminology, workflows, constraints, entity maps, and open questions for later reuse. Also use for long-form or chaptered material such as books, long reports, multi-section drafts, or large corpus updates that require section-aware extraction and synthesis.
Rewrite or draft responses, documents, reviews, explanations, and notes that are too list-heavy, outline-like, or terse. Use when the user asks for fewer bullet points, more natural prose, fuller paragraphs, smoother narrative flow, or when a draft feels choppy, compressed, or over-structured. Typical triggers include turning notes into paragraphs, softening bullet-heavy summaries, expanding terse explanations, and preserving substance while making writing read like connected prose.
Convert a concrete research claim into a tracked, decisive experiment plan that works either as a standalone planning artifact or as the experiment stage inside a coordinated research workflow. Use when asked to design experiments, define baselines or ablations, decide run order, separate must-run from nice-to-have evidence, or turn a claim plus evaluation goal into a validation plan.
Run a stringent, adversarial novelty review over a concrete research idea, method, protocol, artifact, or claimed finding. Use when asked to assess novelty, position a paper or project, build a prior-art matrix, decide whether something is incremental, or pressure-test whether the right move is to proceed, reframe, or abandon. Prefer `research-paper-review` for first-pass technical critique of a single paper artifact and `research-rebuttal` when the task is to answer concrete reviewer comments rather than establish novelty positioning from scratch.
Turn supported claims and evidence into a research paper plan that works either as a standalone manuscript-structuring pass or as the paper stage inside a coordinated research workflow. Use when asked to outline a paper, map claims to evidence, plan figures and citations, align experiments to manuscript claims, or keep limitations and threats to validity explicit.