litreviewlisted

# litreview Turn a topic into a ranked, deduped, honestly-scoped reading list, with a one-paragraph takeaway per paper and a synthesis of what the cluster says. The output is what the user reads before deciding what to cite. ## Boundary with reading-papers This skill starts from a question and ends with a set of papers. The `reading-papers` skill starts from one known paper and ends with its text. If the user hands over a DOI, a link, a title, or an author name and wants that thing read, stop here and use `reading-papers`. The two share machinery. Discovery here is `paper.py search`, reading here is `paper.py get`, and both live in the `reading-papers` skill directory. Read `~/.claude/skills/reading-papers/SKILL.md` before the reading phase if you have not already; it owns the access ladder and the honesty rules that apply to every summary this skill emits. ## Scope the question first A one-word topic gives a one-word answer. Before searching, pin down the actual question and the year window if the user implied one. Ask if the request is bare ("do a lit review"). Do not guess a topic. Scope phrases in the request map onto flags without the user having to type them: "since 2022" is `--since 2022`, "Marketing Science only" is `--venue "Marketing Science"`, "top 30" is `-n 30`, "recent" with no year is the last five years, and say that you read it that way. ## Step 1: search ```bash ~/.claude/skills/reading-papers/scripts/paper.py search "<topic>" -n 25 --json ``` Flag