| name | literature-review |
| description | Write an exploratory literature review based on papers in an autor workspace. For approved review-article workflows, respect the canonical references.bib / reference-map.json / evidence-ledger contract produced by plan. |
Literature Review Writing
Use this skill for open-ended exploratory review writing, research summaries, thesis literature-review chapters, or early state-of-the-art sketches.
For formal review-article production, prefer:
plan to build the canonical planning package
- a user-selected drafting workflow, or
paper-writing when drafting individual sections from that package
Canonical Plan Awareness
If the workspace contains:
references.bib
reference-map.json
review-plan.md
evidence-ledger.md
table-figure-plan.md
then these files are hard constraints, not optional background. Do not rebuild the outline or evidence set during exploratory writing. If the user wants formal prose from that package, use the user's explicitly chosen drafting workflow or paper-writing for specified sections.
Older files such as paper-classification.md, section-evidence.md, table-plan.md, and execution-tasks.md are compatibility exports only. Trust the canonical files when conflicts exist.
Prerequisites
The user should specify a workspace. If not:
- run
autor ws list
- reuse an obvious active workspace or ask the user to choose
Write outputs under workspace/<name>/.
Workflow
1. Clarify the task
Confirm or infer:
- topic or research question
- audience and output type
- language
- approximate length
- whether the task is exploratory or plan-grounded
- whether the canonical planning package already exists
2. Survey the workspace
autor ws show <name>
autor ws search <name> "<topic>"
autor show <dir_name> --level 2
Escalate core papers to L3/L4 only when needed:
autor show <dir_name> --level 3
autor show <dir_name> --level 4
L3 is a paper-level conclusion card. It may include an inferred synthesis when a paper lacks a clear conclusion section. Use it for orientation, but verify numerical or controversial claims against L4 or an approved evidence ledger.
Use citation graph tools for relationship checks:
autor refs "<id>"
autor citing "<id>"
autor shared-refs "<id1>" "<id2>"
3. Structure the review
If no canonical plan exists, propose an outline based on the workspace and user question.
If a canonical plan exists:
- follow
review-plan.md
- use evidence from
evidence-ledger.md
- use citation keys from
references.bib
- respect roles and full-text status in
reference-map.json
- use
table-figure-plan.md for tables and figures
When the plan conflicts with new reading, return to plan instead of silently reshuffling the draft.
4. Draft
Writing principles:
- synthesize rather than enumerate papers
- identify contradictions, limitations, and boundary conditions
- distinguish authors' claims from established facts
- use citation keys from
references.bib when the canonical package exists
- for informal early drafts without a canonical package, still avoid hallucinated citations and export references when possible
If drafting from an approved plan, do not rebuild the plan or evidence set. Draft only under the user's explicitly chosen workflow, or use paper-writing for a specified section.
5. Save and check
Default output:
workspace/<name>/literature-review.md
If citations are needed:
autor ws export <name> -o workspace/<name>/references.bib
Before delivery, check that:
- the text matches the requested scope
- citations are real and local to the workspace
- any deviation from a canonical plan is explicitly justified or routed back to
plan
Academic Attitude
- Paper conclusions are claims, not truth.
- High citation counts do not equal correctness.
- When papers disagree, explain possible causes rather than smoothing the disagreement away.
- Avoid turning review articles into substitutes for primary evidence.