| name | literature-review-drafting |
| description | Use when drafting or revising paper-facing literature-review prose from already-collected research records, especially background sections that need inline source citations for substantive claims. |
Literature Review Drafting
Draft narrative literature-review prose from published evidence records. Treat this as a writing layer over the evidence graph, not as new evidence extraction.
Read First
lit-review-studio.config.json and the active LIT_REVIEW_STUDIO_DOMAIN value.
domain-packs/<domain-id>/domain.json
domain-packs/<domain-id>/taxonomy.v1.json
domain-packs/<domain-id>/public-copy.v1.json
domain-packs/<domain-id>/skills/literature-review-drafting.md when present.
- Relevant published records under
data/sources, data/findings, data/claims, data/artifacts, data/evidence-appraisals, and data/publication-events.
- Existing synthesis reports under
research/syntheses/ when they summarize the target scope.
Workflow
- Resolve the intended paper audience, target section, domain, and scope. If the user has not specified length or style, write a concise background-section draft.
- Inspect existing
report_artifact records for a current literature_review with the same domain and paper-facing scope. Revise that report by default; create a new current literature review only when the audience, target section, or scope is intentionally different, or when the old report is marked superseded or archived.
- Inventory the published sources, findings, claims, and limitations that can support the requested prose. Treat sidecar syntheses as traceability aids, not as substitutes for the underlying published records.
- Build a claim-to-citation map before drafting. Every substantive factual claim should cite one or more concrete source records inline.
- Separate evidence types in the prose:
- empirical study findings,
- standards or policy guidance,
- adjacent evidence,
- the author's interpretation or implication for the present study.
- Draft or revise a Markdown literature review with readable narrative paragraphs, not an audit table. Use author-year citation text when source metadata supports it, and retain source IDs where traceability matters.
- Include a short limitations or gaps paragraph that states what the reviewed literature does not establish.
- Add a traceability note listing the source IDs used, unless the user explicitly asks for polished prose only.
- Create or update a
report_artifact record under data/report-artifacts/ when the draft should appear in the app. Include artifact_type: "literature_review", status, domain_id, scope_ids, path, source_ids, claim_ids, and citation-audit status. When replacing a current literature review, mark the older artifact superseded or archived.
- Run validation and tests when skill files, domain adapters, report indexes, or repo-observed behavior changed:
npm run validate
npm test
Boundaries
- Do not browse for new sources unless the user asks for new literature search or current-source verification.
- Do not create or publish new source, finding, claim, or artifact records from a literature-review pass unless the user explicitly requests a candidate bundle.
- Do not cite a claim without citing the underlying source record when writing paper-facing prose.
- Do not present targeted baseline review coverage as systematic-review coverage.
- Do not turn standards, guidance, or policy documents into empirical findings.
- Do not use source IDs as a substitute for readable citations when the user wants paper-facing prose; use IDs as traceability support.
- Do not make operational recommendations stronger than the published evidence supports.
- Do not leave multiple current
literature_review artifacts for the same domain and paper-facing scope as an accidental result of additive writing.
- Do not promote a sidecar synthesis into paper-facing prose until its underlying evidence is published or its unpublished status is explicitly bounded in the prose.
Expected Outputs
- One Markdown literature-review draft or revision under
research/syntheses/ unless the user specifies another location.
- One current
report_artifact index record when the draft should be visible in the app, with older replaced literature reviews marked superseded or archived.
- Inline citations for every substantive factual claim.
- A short limitations/gaps paragraph.
- A traceability section listing cited source IDs and, when useful, supporting claim or finding IDs.
- Passing validation and tests when repository skills, adapters, or observed behavior are changed.