zjunlp / sciatlas-literature-review

SciAtlas Literature Review

Use this skill to run the repository literature-review workflow. The workflow performs topic profiling, SciAtlas backend paper search, evidence organization, method clustering, time slicing, outline planning, evidence-pack construction, and optionally full section drafting/integration.

Operating Contract

• Run only sciatlas literature-review or python run_sciatlas.py literature-review for this skill.
• Own the end-to-end novice flow: install or locate the CLI, guide registration, configure .env or shell variables, run the workflow, inspect artifacts, and synthesize the final review result.
• Ask the user only for human-only values: missing topic/domain, email, verification code, SciAtlas token, LLM credentials that are not already configured, or one necessary scope clarification.
• Do not ask the user to run shell commands when tool access is available.
• Use --workflow flash by default for interactive work.
• Use --workflow full when the user requests a comprehensive formal review or when flash artifacts are too thin.
• Never disclose full API keys or tokens.
• Read saved artifacts before answering.

Zero-Start Bootstrap

1. Check whether the repository command works:

python run_sciatlas.py literature-review -h

If needed, fall back to sciatlas literature-review -h after installing the full checkout.

1. This dedicated workflow requires a full SciAtlas checkout. If it is missing, clone the repository, change into it, then run python -m pip install -e ./sciatlas and python -m pip install -r requirements-workflows.txt. Do not use the GitHub #subdirectory=sciatlas package-only installation for this workflow.
2. Check current environment and .env for SCIATLAS_API_KEY and LLM settings before asking the user.
3. If no SciAtlas token is configured, guide the user to http://sciatlas.openkg.cn/register; ask for email, verification code, and returned sciatlas_xxx token only when needed.
4. If LLM credentials are required and missing, ask only for the missing values. Use the user's provider values without printing them back.
5. Configure the current shell or .env yourself, then run the workflow.

Paper retrieval for this workflow must use only the provided SciAtlas backend through run_sciatlas.py search-papers. Do not add public-paper fallback retrieval (Semantic Scholar, OpenAlex, Crossref, arXiv scraping, or browser search) when the SciAtlas backend returns no papers or errors. The local literature-review code may still use LLM calls for planning, clustering, and synthesis.

Configure workflow credentials in .env or the shell:

SCIATLAS_API_BASE_URL=http://sciatlas.openkg.cn
SCIATLAS_API_KEY=<sciatlas-token>
OPENAI_API_KEY=<llm-key>
OPENAI_BASE_URL=https://api.deepseek.com
LLM_MODEL=deepseek-v4-flash

The workflow also accepts DMX-API-KEY, DMX_API_KEY, LLM_API_KEY, LLM_BASE_URL, LLM_API_URL, SEARCH_LLM_API_KEY, SEARCH_LLM_API_URL, and SEARCH_LLM_MODEL.

Run Plan

Flash path:

python run_sciatlas.py literature-review \
  --query "<topic>" \
  --domain "<optional field>" \
  --workflow flash

Full path:

python run_sciatlas.py literature-review \
  --query "<topic>" \
  --domain "<optional field>" \
  --workflow full

Useful overrides:

• --top-k N changes the default SciAtlas search budget.
• --probe-top-k N and --round-top-k N tune search breadth.
• --round1-action-limit N and --round2-action-limit N tune query planning.
• --report-stop-after outline|packs|full controls report generation depth.
• --subject-domain general|chemistry|biology selects prompt constraints.
• --smoke runs the mock search path for structure validation.

Smooth flash defaults when the backend is slow or unstable:

python run_sciatlas.py literature-review \
  --query "<topic>" \
  --domain "<optional field>" \
  --workflow flash \
  --probe-top-k 3 \
  --round-top-k 3 \
  --round1-action-limit 1 \
  --llm-paper-limit 12 \
  --report-stop-after packs \
  --workflow-timeout 600 \
  --llm-timeout 120 \
  --outline-timeout 180

Set SCIATLAS_SEARCH_TIMEOUT for the per-action hosted search-papers timeout. If SciAtlas returns 5xx, keep the error in artifacts and do not switch to public retrieval.

Workflow Modes

flash compresses nonessential stages:

• smaller probe and round search budgets;
• fewer Round 1 actions and no Round 2 by default;
• query cleaning and relevance guard enabled;
• report generation stops after evidence packs, producing a fast outline/evidence report.

full runs the broader path:

• larger probe and round search budgets;
• Round 2 refinement enabled;
• KG policy, query cleaning, and relevance guard enabled;
• full formal review drafting and integration.

Artifacts To Read

Read the run directory:

• summary.json: status, workflow mode, subprocess logs, and artifact pointers.
• report.md: user-facing review, outline/evidence-pack summary, or full formal review.
• lr_search/search_result.json: topic profile, time windows, paper cards, clusters, coverage report, and search actions.
• lr_search/organized_search_result.json: deterministic evidence map when generated.
• lr_review/formal_outline.json: planned review structure.
• lr_review/citation_plan.json, section_packs/, subsection_packs/: flash evidence packs.
• lr_review/formal_review.md and diagnostic_report.md: full-mode final outputs.
• logs/*.txt: subprocess stdout/stderr when a stage fails.

In flash, a report that stops after packs is normal success. Do not rerun full unless the user asks or the evidence is insufficient for their requested depth.

Deliverable

Return:

• exact command used, with credentials omitted;
• workflow mode and artifact paths;
• concise topic scope;
• 5-12 representative papers or paper groups;
• method/theme clusters and timeline notes;
• review outline or full review summary;
• gaps, caveats, and next SciAtlas queries.

Keep paper claims tied to artifact titles, clusters, and evidence fields.