Delegate Compound Engineering ideation, brainstorming, and planning workflows to an interactive Claude Code session from Codex. Use when Codex should stay the outer operator while Claude Code runs CE slash skills, preserves native question prompts, and writes durable local artifacts.
Delegate a general task from Codex to a normal interactive Claude Code session in a target project. Use when Codex should stay the outer operator while Claude Code uses native skills or subagents, writes durable artifacts, and asks questions through Codex.
Delegate the current PR branch to Claude Code and Compound Engineering for code review, post GitHub PR review comments, then let Codex fix actionable findings. Use when a PR needs an external Claude Code review loop with durable artifacts and guarded GitHub posting.
Draft, critique, refine, or activate evidence-checkable agent Goals from rough tasks, brainstorm requirements, implementation plans, grill-with-docs decisions, or existing weak /goal text. Use when the user asks to create a Codex or Claude Code goal, turn a plan or brainstorm into a goal, strengthen a goal, decide…
Finish one obvious coherent change through a thin GitHub landing lifecycle: preserve or create the feature branch and commits, reuse or run validation once, push, create or reuse a pull request, wait only for required checks, merge the exact head, verify GitHub's result, and safely synchronize the primary checkout's…
Evidence-first disk cleanup and external-drive storage decisions for macOS/Linux. Use when the user asks what is using disk space, how to free hard-drive space, what can be deleted safely, what should be archived to an external drive, or how to build a cleanup decision log.
Use the @steipete/oracle CLI to bundle a prompt plus the right files for ChatGPT review. Render/copy with Oracle, then submit through the user's Chrome session by default when safe: prefer the @Chrome plugin, then Computer Use in Chrome, then @Browser as a fallback. Use the latest Pro model with extended thinking when…
Read, summarize, search, and extract structured artifacts from PDF files, especially papers and reports. Use when the user wants a PDF converted before reading, needs figures or tables pulled out, wants image labels or captions, or needs cleaner Markdown/text than reading the PDF directly.
Personalized Python tutoring with persistent memory, daily Q/A logs, and dynamic level adaptation. Use when users want to learn Python progressively across sessions, track strengths and gaps, or receive explanations and practice matched to their current ability.