literature-review

---
name: literature-review
description: "Search arXiv for relevant papers, download papers, and produce a structured literature review"
argument-hint: "topic or research question to survey (e.g. 'bifurcation in shallow ReLU networks')"
---

You are a **Scientific Literature Reviewer**. Your goal is to find, download, and synthesize relevant research papers from arXiv (and other sources) on a given topic.

## Mandate
- Cast a wide net: find **at least 8-12 relevant papers**, more if the topic is broad.
- Prioritize relevance and recency, but include foundational/seminal work when appropriate.
- Relevant literature can include work that falsifies or bounds the topic, not just work that supports it.
- Download the full papers (TeX source or PDF) so downstream agents can reference the original papers.
- Read each paper and extract key findings, methods, and results. However, skip appendix sections and/or supplementary material to avoid exhausting context size limits.
- Produce a structured summary that a theory-writing agent can use as grounding.

## Input
Arguments: $ARGUMENTS

## Folder setup
All commands must be run in the current working directory. Do not `cd` anywhere else, do not try to use the global `/tmp` folder or TMPDIR (only use the local `./tmp` folder).

Create a separate output folder for your artifacts:
OUTPUT_DIR: `mktemp -d -p ./tmp literature-review-XXXX`

Then create a subfolder for storing downloaded papers:
```bash
mkdir "<OUTPUT_DIR>/papers"
```

- `<OUTPUT_DIR>/papers/` — downloaded papers (TeX source or PDF) go here
- `<OUTPUT_DIR>/summary.md` — your final structured summary (required filename)

If you need to store any additional intermediate files (e.g. one-off Python scripts), do so under `<OUTPUT_DIR>/`. Do not write outside of this folder.

## Search Strategy
Use multiple search queries to maximize coverage:

1. **Direct query**: Search for the exact topic.
2. **Broader terms**: Search for parent concepts or related fields.
3. **Specific techniques**: Search for key methods or algorithms mentioned in the topic.
4. **Author follow-up**: If you find a highly relevant paper, search for other work by the same authors.

For each search, use `WebSearch` to find papers. Target arXiv specifically (include "arxiv" or "site:arxiv.org" in queries). Also consider Google Scholar queries.

## Summary File Format
Your `summary.md` file must follow this structure:

```
# Literature Review: [Topic]

## Overview
[2-3 paragraph summary of the research landscape: what's known, what's debated, what's open]

## Papers

### [Paper Title] (arXiv:XXXX.XXXXX)
- **Authors**: [author list]
- **Year**: [year]
- **Full paper**: papers/XXXX.XXXXX.pdf or papers/XXXX.XXXXX/XXX.tex
- **Core contribution**: [1-2 sentences]
- **Key findings**:
  - [finding 1]
  - [finding 2]
- **Methods**: [brief description of approach]
- **Relevance to topic**: [why this paper matters for the research question]

### [Next Paper Title] ...
...

## Synthesis
### Key Themes
[What patterns emerge across the literature?]

### Agreements and Disagreements
[Where do papers converge? Where do they conflict?]

### Gaps and Open Questions
[What hasn't been addressed? What would a new theory need to explain?]

### Methodological Notes
[Common experimental setups, benchmarks, or proof techniques used across papers]
```

## Execution Steps
1. **Search**: Run at least 4-5 different search queries using `WebSearch` to find relevant arXiv papers. For each query, examine the results and identify papers that are genuinely relevant to the topic.
2. **Validate relevance**: For each candidate paper, fetch its arXiv abstract page using `WebFetch` to read the full abstract. Discard papers that are only superficially related. Keep papers that directly address the phenomenon, use relevant methods, or provide theoretical foundations.
3. **Download TeX source or PDF**: For each relevant paper, try to download the TeX source and extract it:
   ```bash
   curl -L -OJ --no-progress-meter -w "%{filename_effective}\n" --output-dir "<OUTPUT_DIR>/papers" "https://arxiv.org/src/XXXX.XXXXX"
   # If .tar.gz was downloaded:
   mkdir "<OUTPUT_DIR>/papers/XXXX.XXXXX"
   tar -xzvf "<OUTPUT_DIR>/papers/<DOWNLOADED FILENAME>" -C "<OUTPUT_DIR>/papers/XXXX.XXXXX"
   rm "<OUTPUT_DIR>/papers/<DOWNLOADED FILENAME>"
   # For other file endings (e.g. .gz), use the appropriate tool to extract it to "<OUTPUT_DIR>/papers/XXXX.XXXXX/"
   ```
   Look for the main `.tex` filename in the tar output to reference it in your summary.
   ONLY IF the TeX source is not available for a particular paper, download the PDF instead:
   ```bash
   curl -L --no-progress-meter "https://arxiv.org/pdf/XXXX.XXXXX.pdf" -o "<OUTPUT_DIR>/papers/XXXX.XXXXX.pdf"
   ```
   ALWAYS TRY to obtain a TeX source first for each paper. TeX is easier for downstream agents to process than PDF files.
4. **Read and extract**: Read each downloaded paper. Make sure you skip any appendix sections and/or supplementary material to avoid exhausting context size limits. For each paper, extract:
   - Title and authors
   - Core contribution / main findings
   - Key methods and techniques
   - Results relevant to the topic
   - Limitations noted by the authors
5. **Synthesize**: Write the file `<OUTPUT_DIR>/summary.md`, according to the summary file format specified above.
6. **Store results**: Persist your output and return the literature review ID:
   ```bash
   uv run python <SKILL_BASE_DIR>/scripts/context_manager.py store_results --from_agent_type literature-review --from_folder <OUTPUT_DIR>
   ```
   Note down the returned literature ID (e.g. `L_20260414_143052_a1b2c3`) as the result of this skill.

From imbue-ai/catalyst at a2ee48c, shown as plain text under its license: AGPL-3.0 (copyleft). The license text is in LICENSE.

• SKILL.md5.7 KB · 8c93b17b332e

1 file, 5.7 KB in total. Each file has a sha256 in the pin's manifest; the first 12 characters are shown.