| name | literature-review |
| description | Search and review ecosystem and biogeochemical literature using MCP tools (PubMed, arXiv, bioRxiv, Google Scholar), with optional full-text reading and citation snowballing. Use when the user needs evidence, citations, methods comparison, or novelty checks for EcoSIM and warming-nitrogen hypotheses. |
| allowed-tools | Bash, Read, Write, WebSearch, Agent |
| user-invocable | true |
Literature Review Skill
Search, read, and synthesize literature relevant to EcoSIM modeling, warming experiments, nitrogen cycling, and related ecosystem biogeochemistry.
Compatibility Modes
- Full mode: MCP servers available (
pubmed, paper-search) for richer search and full text.
- Limited mode: WebSearch + DOI fallback with explicit reduced-coverage note.
Prerequisites
Preferred servers are configured in .claude/.mcp.json:
pubmed for rich biomedical search and PMC full text
paper-search for arXiv, bioRxiv, medRxiv, and Google Scholar search/read
If MCP servers are unavailable, fall back to WebSearch and DOI landing pages.
Scope
Prioritize evidence for:
- Experimental warming impacts on nitrogen processes
- Soil moisture-temperature interactions
- N2O emissions, nitrification, denitrification, mineralization
- EcoSIM-relevant forcing/parameterization methods
- Ecosystem-specific effects (tundra, forest, grassland, agricultural)
Review Tiers
| Tier | Papers | Full text | Citation expansion | When to use |
|---|
| Quick scan | 5-10 | No | No | Fast orientation |
| Standard review | 20-30 | Top 10 | Yes | Default for project work |
| Deep review | 50+ | Top 20 | Yes | Comprehensive synthesis |
Default: quick for ad-hoc questions, standard for project work, deep only when explicitly requested.
Workflow
Step 1: Clarify the Question
Capture:
- Topic and hypothesis
- Organisms/ecosystems
- Time horizon (recent vs historical)
- Whether methods comparison is required
Step 2: Build Query Set
Construct 3 query families:
- Core process query (for example: warming + nitrification)
- Ecosystem query (for example: tundra + soil nitrogen)
- Methods query (for example: incubation, isotope tracing, flux chamber)
Step 3: Discover, Deduplicate, Rank
Search across PubMed, bioRxiv, arXiv, and Google Scholar.
Deduplicate by DOI/PMID.
Rank by EcoSIM relevance:
- HIGH: directly measures N-cycling or climate-soil controls used in EcoSIM workflows
- MEDIUM: adjacent mechanisms or methods transferable to EcoSIM
- LOW: tangential background
Step 4: Optional Deep Reading
For standard/deep tiers, read top papers in full text where available.
Extract:
- Study design and sample sizes
- Quantitative outcomes (effect sizes, p-values, uncertainty)
- Limitations and caveats
- Variables that map to EcoSIM or BERVO terms
Step 5: Summarize
Produce:
- Thematic findings
- Methods comparison table
- Key quantitative results table
- Evidence quality notes
- Knowledge gaps and contested findings
Use this structure:
## Literature Review: <topic>
Review depth: <quick|standard|deep>
Papers found: <N>
Full text read: <N>
Sources searched: <list>
### Summary
<2-3 sentence synthesis>
### Key Findings by Theme
- <Author Year>: <finding> (DOI/PMID)
### Methods Comparison
| Study | Ecosystem | Method | Sample size | Key metric |
|---|---|---|---|---|
### Key Quantitative Results
| Finding | Value | Study | Evidence quality |
|---|---|---|---|
### Gaps in Current Knowledge
- <gap>
### Relevance to EcoSIM
- <variables, parameters, forcing constraints, benchmark opportunities>
Step 6: Store References
If a project location is known, append references to a repo-local file such as:
docs/references/<topic>.md or
derived/<study-id>/summary.md (when extending an existing study packet)
Do not require a BERIL projects/{id} layout.
Integration
- Called by:
suggest-research, synthesize
- Inputs from:
derived/*.tsv, study summaries, user hypotheses
- Outputs: a citation-backed review and optional reference file updates
Fallback
If MCP is unavailable:
- Use WebSearch with source constraints (
site:pubmed.ncbi.nlm.nih.gov, DOI URLs)
- Note reduced coverage and lack of robust full-text retrieval