skills

Use when starting feature work that needs isolation, before executing implementation plans, or whenever the user mentions branching, parallel work, or wanting a clean workspace. Creates isolated git worktrees with systematic setup and safety checks. Use this skill even if the user just says "let's work on this in a branch" or "set up a workspace for this."

Read an academic paper (PDF) end to end, explain it intuitively to the reviewer, critique it honestly, and produce a submittable peer review. Use this whenever the user wants to review, referee, critique, assess, or "understand then evaluate" a paper or manuscript — conference/journal reviews, referee reports, meta-reviews, or requests like "explain this paper then write a review", "do an MLHC/NeurIPS/ICLR/CVPR review", "help me review submission N". Trigger even if the user only says "take a look at this paper" in a reviewing context, or points at a PDF plus a rubric/README.

Iterate on complex, multi-file implementation plans through investigation, drafting, review, and refinement cycles. Use this skill when the user wants to create or improve a plan for a multi-step project, review an existing plan for gaps, conduct an investigation to inform a plan, decompose a large task into chunks with sub-plans, or do a systematic review/feedback cycle on plan documents. Also use when the user seems stuck on what to do next with a plan, or when they want to validate assumptions before building. Trigger whenever the user mentions "plan", "sub-plan", "review the plan", "what are we missing", "iterate on this", "what should we investigate", or wants to go from vague goals to a concrete implementation roadmap.

Use this skill when refining Codex approval behavior, sandbox settings, or allow-rules. It inspects local Codex config, rules, and session history to find recurring approval friction, separate rule issues from sandbox/cache issues, and propose generalizable improvements instead of one-off exceptions.