
Audit AI-generated or AI-shaped backend/general code diffs for duplicate helpers, over-defensive control flow, broad exception wrappers, speculative scaffolding, comment/docstring boilerplate, local style drift, hallucinated APIs/dependencies, fixture-shaped test hacks, and obvious safety/performance gaps. Use when reviewing or safely cleaning up Python, TypeScript, or other implementation code after a feature, bugfix, or prototype pass.
Reference-only checklist for AI-writing artifacts, citation failures, and cleanup rewrites in Markdown, MDX, wiki text, or pasted chatbot output. Use when you need objective residue checks, false-positive-safe prose triage, and practical fixes without relying on detector scores.
Visual-reference-to-app implementation loop for building or restyling a web app screen to match a provided reference image. Use when Codex is asked to take a screenshot/reference/mockup/image and build an app or route that looks like it, especially when the task requires hosting the app, capturing Playwright screenshots, comparing against the reference, and repeatedly editing until the result is close.
Codex-specific, session-driven self-improvement for Codex behavior and project instructions. Use when the user asks to inspect past Codex sessions, run a "dream" pass over prior interactions, mine repeated user corrections/preferences, improve or draft skills, update repo/project `AGENTS.md` guidance, or propose durable edits to global `~/.codex/AGENTS.md`.
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.
OCR PDFs with docling while tracking per-page text and rasterize PDFs to images. Use for PDF ingestion, page-aware text extraction, rendering pages to images, or inspecting PDF metadata, with outputs saved under a local project directory.
Use when a user asks to debug or fix failing GitHub PR checks that run in GitHub Actions; use `oai_gh` or `gh` to inspect checks and logs, summarize failure context, draft a fix plan, and implement only after explicit approval. Treat external providers (for example Buildkite) as out of scope and report only the details URL.
Help address review/issue comments on the open GitHub PR for the current branch using `oai_gh` or `gh`; verify auth first and prompt the user to authenticate if not logged in.
Review git changes and split them into semantic commits with clear messages. Use when the user asks to commit work, clean up local history, or group a mixed diff into logical commits. Do not commit on main or master unless the user explicitly asks.
Use when Codex needs to write, rewrite, critique, or reply on Twitter/X in Jason Liu's personal voice. Trigger for requests like "tweet like me", "write this in my style", "make this sound like Jason", "draft a reply", or when Jason asks for Twitter copy about Codex, product building, feedback, launches, quote-tweets, or operator/value takes.
Create or edit Slidev presentations in the /Users/jasonliu/dev/presentations repo. Use for drafting new decks, editing existing slides, applying repo-specific Slidev conventions, and polishing/animation work. Triggers: Slidev slide requests, layout/components usage, deck setup, or presentation workflow guidance for this repo.
Audit and clean Codex-created Git worktrees and leftover worktree directories safely. Use when disk usage appears inflated by ~/.codex/worktrees, Git worktree metadata, detached Codex worktrees, stale branch worktrees, or when the user asks to map worktrees to pull requests before deletion.
Draft, rewrite, or critique executive-facing communication so it is answer-first, concise, evidence-backed, audience-specific, and explicit about decisions or asks. Use for leadership updates, launch/status readouts, board or staff prep, escalation notes, Slack/email updates to executives, PR/FAQ summaries, weekly business updates, and turning messy project context into a clear executive narrative.
Build or refine single-file information-first HTML artifacts, especially index.html or text.html pages, with strong information hierarchy, restrained styling, accessible semantics, and minimal AI-generated frontend tells. Use when creating static HTML reports, research pages, explainers, briefs, dashboards, note indexes, or simple front ends whose goal is comprehension rather than marketing conversion.
Use only when the user explicitly asks to stage, commit, push, and open a GitHub pull request in one flow using the GitHub CLI (`gh`).
Extract transcripts, titles, and thumbnails from YouTube videos. Use for ingesting video content, capturing captions with timestamps, or downloading video metadata.
Delegate codebase exploration or scripted actions to a non-interactive Codex exec run (codex exec / codex e). Use when you want a subagent to read lots of code or take actions without human interaction, and you can accept CLI output and optional file changes as the result.
Audit AI-generated or AI-looking frontend implementations, UI screenshots, and design diffs for generic AI aesthetics, card/gradient/font tells, weak UX copy, accessibility gaps, brittle responsive behavior, and one-off design-system drift. Use when reviewing or restyling React, Tailwind, shadcn/ui, HTML/CSS, landing pages, dashboards, or app screens to make the UI feel more intentional without a full redesign unless explicitly requested.
Use when tasks involve reading, creating, or reviewing PDF files where rendering and layout matter; prefer visual checks by rendering pages (Poppler) and use Python tools such as `reportlab`, `pdfplumber`, and `pypdf` for generation and extraction.