| name | lit-review |
| description | Build or extend a literature survey on a research area — search, triage, map the field, and identify open problems. Use for "/lit-review <area>", building a reading queue, or finding the research delta for a proposal. |
Literature Review
Build a literature map for the area in the arguments, producing a survey document under research/reading-lists/ (queue + map) and, when requested, a full written survey under research/.
Building the map
- Seed — start from papers the user already has in
research/paper-notes/, the seminal-papers lists in the relevant curriculum/ module, and (if web access is available) recent surveys and the citation graph around them.
- Triage — for each candidate paper record one line: claim, method family, venue/year, and why it's in or out. Target a queue of 20–50 papers organized into 3–6 thematic clusters, not a flat list.
- Map structure — for each cluster identify: the founding paper, the current state of the art, the standing disagreement (every live field has one), and what evidence would settle it.
- Delta hunting — the survey's job is to end in open problems. For each cluster, state: what's underexplored, why (hard? unfashionable? recently unblocked?), and a feasible 12-week problem slice per the Phase 3 criteria in
milestones/roadmap.md.
Quality bar
- Every claim about a paper must be checkable — cite the specific paper, never "studies show".
- Coverage honesty: explicitly list what the survey does not cover and why the boundary is drawn there.
- Recency check: if the newest paper in a cluster is >2 years old, either justify the cluster as settled or flag the search as incomplete.
Outputs
research/reading-lists/<area>.md — clustered, annotated queue with priorities.
- Each deep-read paper gets its own note via the
/paper workflow.
- Full survey (when requested): problem framing → cluster-by-cluster synthesis → comparison table → open problems ranked by value/feasibility. Have the
skeptical-reviewer agent attack the draft before calling it done.