codex-toolbox

Use when a user asks to set up, create, prepare, or start a brief Obsidian PaperRead draft or paper-reading note.

Use when a user asks to review, critique, fact-check, strengthen, or annotate an existing Obsidian PaperRead or paper-reading note by adding source-backed comments inside that note.

Use when a user wants to read current articles from configured WeChat subscriptions, configure selected sources, or receive an incremental scheduled digest through BestBlogs.

Find, check, classify, add, or repair one academic paper across public scholarly sources and the user's Zotero Research library. Use for a title, DOI, arXiv ID/URL, or publisher URL when the user asks whether the paper is saved, where it belongs, or asks to add/save/import it with a verified PDF and Research/ReadLater filing.

Maintain a Karpathy-style, source-backed research LLM Wiki in the user's Obsidian vault. Use when the user asks to ingest papers, Zotero items, web clips, PDFs, or paper notes into `Research/LLM Wiki`; query or synthesize from that wiki; file durable research analyses; lint the wiki for stale claims, missing citations, orphan pages, duplicate concepts, or contradictions; or mentions `$research-llm-wiki ingest`, `$research-llm-wiki query`, or `$research-llm-wiki lint`.

Use when the user asks to explain, teach, clarify, compare, walk through, or help them understand a concept, system behavior, error, or code; especially for why/how questions or when a definition alone is not enough. Do not use for execution-only requests or simple facts that need only a terse answer.

Prepare a concise, strictly read-only daily command-center brief across Gmail, Google Calendar, and Todoist. Use for daily or morning briefs, daily planning, and scheduled command-center runs that need email context, time commitments, and priority tasks without changing any source.

Use when Codex is in Plan Mode for non-trivial, ambiguous, architectural, multi-step, high-risk, or orchestration-bound work; when reviewing a plan before coding; or when the user asks to think deeply, challenge assumptions, do adversarial planning, or choose between Codex-only, native Codex subagents, Superpowers, and OpenSpec. Do not use for tiny edits, simple command-output checks, pure execution, post-code verification, or full Superpowers design-doc workflows.

Manage a personal task system across Todoist and Google Calendar. Use when the user asks to add, edit, complete, find, review, or organize tasks; capture deadlines or task details; block focused work time; or schedule a remote meeting while keeping tasks and calendar events distinct.

Extract one local PDF, image, DOCX, PPTX, or XLSX with a quality-first MinerU workflow and a checksum-verified manifest. Use for complex, scanned, OCR-heavy, table/formula-rich, or layout-sensitive documents where simple text extraction may lose structure, and for bounded page-range extraction that must preserve auditable local artifacts.

Use when Codex needs to create, review, or repair reproducible AI/systems paper figure workflows with editable draw.io or diagrams.net pipeline diagrams, Matplotlib and SciencePlots experimental plots, SVG/PDF exports, Inkscape cleanup, figures_src/, figures/, Makefile targets, or publication-quality vector figures.

Use when finance workflows need Vibe-Trading market data, backtests, factor analysis, screening, research swarms, Shadow Account reports, or MCP tool discovery.

Use only for PixelLab pixel-art game asset workflows, including sprites, character rotations, animations, top-down or sidescroller tilesets, isometric tiles, and map objects. Do not use for generic image generation, web search, local file work, or normal coding unless the user explicitly wants PixelLab assets.

Use when the task requires automating a real browser from the terminal (navigation, form filling, snapshots, screenshots, data extraction, UI-flow debugging) via `playwright-cli` or the bundled wrapper script.