Context-aware router that detects work type (papers, proposals, code, figures, admin) and dispatches to the right domain skill. Works within superpowers' execution phase as the inner decision loop.
Use when auditing a paper's existing .bib for hallucinated citations or stale metadata before submission, especially when bib entries came from co-authors or automated tools and may include fabricated authors, future arXiv IDs, or preprints already published in peer-reviewed venues. Companion to bibref-filler (which adds new citations); this skill audits existing ones.
Build copy-ready prompts for downstream LLM figure generation from source text, figure ideas, prior examples, or rough drafts. Use when Codex needs to decide whether a section needs a figure, distill the figure's core message and required visual elements, draft a model-neutral prompt pack for image-generation models, adapt prompts to a requested tool, or recommend limited cleanup after external generation.
Compose NSF proposal outlines and section drafts from a project idea, prior proposal examples, and solicitation requirements. Use when Codex needs to choose between ecosystem-building and methodology-driven proposal families, turn a brief into a proposal structure, draft or refresh LaTeX sections, introduce an optional cross-section running example for abstract infrastructure proposals, or keep summary, aims, evaluation, and impact sections aligned during proposal writing.
Unified NSF proposal requirements, question-answering, and preflight checking for a single proposal repository. Use when a user has an NSF solicitation PDF/text and needs to (1) ingest and normalize proposal requirements, (2) ask drafting-time questions such as "do we need X?" or "where does this go?", or (3) run a pre-submission gap check against the current proposal files.
Refine one or more NSF thrust, aim, or study files after the initial scaffold exists. Use when Codex needs to deepen a core proposal unit with sharper gaps, local literature grounding, concrete technical measures, evaluation hooks, reviewer-facing running examples or artifact sketches, figure ideas, or visible TODO placeholders while preserving alignment with the current intro, summary, and evaluation plan.
Create 1-page or 2-page CV variants from a longer master LaTeX CV. Use when the user asks for a short CV, concise CV, one-page CV, two-page CV, resume-style version, or a role-specific application CV derived from `Latex/sample_cv.tex` or another master CV source.
Draft grounded area chair, senior area chair, or editor style meta reviews for computer science and machine learning papers from the paper PDF, external reviews, reviewer discussion, rebuttal, and venue decision scale. Use when Codex needs to synthesize reviewer consensus, summarize rebuttal impact, recommend a final decision, or write venue-specific meta reviews for venues such as TNNLS, ICML, NeurIPS, or similar.