Claude Scholar
Semi-automated research assistant for academic research and software development, supporting multiple AI CLI backends.
Evidence shows the installer backs up overwritten files and preserves existing configs (e.g., API keys, permissions), indicating intent for least privilege and user confirmation. However, no specific permission lists or data flow descriptions are provided, so data flow transparency and sensitive data handling score low. Dependency security lacks audits or lock files. External effects: installer modifies user config but provides uninstall and backups, scoring moderate. Rollback exists via backups and uninstall, but untested. Source attribution: no explicit content provenance, but external tools are referenced.
Self-consistency is good; README aligns with workflow descriptions. Dependency availability: relies on external tools (Zotero, Obsidian) without version pinning or mirrors. Failure messages: no examples of error handling or user prompts.
Target audience and scenarios are clear, capability boundaries are explicit (semi-automated, human decisions). Trigger precision: commands and skills have clear triggers. Environment fit: supports multiple platforms and CLIs, but lacks detailed configuration requirements.
Information architecture is clear with quick navigation and doc links. Install notes are detailed with multiple options. Naming stability: version history present. Examples and FAQ: usage scenarios provided, but no FAQ. Known limitations: only some mentioned. License is MIT, versioning has changelog. Maintenance responsibility: no explicit maintainer or contribution guide.
Output usability: typical outputs are described. Marginal value: automation workflows provided, but no comparison with alternatives. Cost-benefit: no performance or resource data.
Claim traceability: workflows linked to evidence records. Cross-source corroboration: no external validation. Fact-inference separation: clearly distinguishes evidence and inference.
- Installer modifies user config; test in isolated environment.
- External dependencies (Zotero, Obsidian) are not version-pinned, affecting reproducibility.
- No security audit or dependency vulnerability scanning provided.
What does this agent do, and when should you use it?
Claude Scholar is a semi-automated research assistant designed for computer science and AI researchers, supporting Claude Code, Codex CLI, Kimi Code CLI, and OpenCode environments. It is not a fully autonomous scientist; instead, it keeps human decisions at the center and automates heavy, repetitive tasks like literature organization, experiment analysis, reporting, and paper writing. The repository ships a rich set of skills (e.g., research-ideation, results-analysis, nature-writing), agents (e.g., literature-reviewer, paper-miner), and slash commands (e.g., /research-init, /rebuttal). The installer deploys these components to the user's ~/.claude directory, and optional integrations with Zotero and Obsidian enable literature management and a durable knowledge base.
Claude Scholar installs skills, commands, agents, rules, and hooks via scripts/setup.sh, copying them into ~/.claude/. Its core workflow is question → evidence → experiment → analysis → claim → writing. Concrete operations include: using research-ideation to turn vague topics into structured questions; importing and organizing papers through Zotero MCP; converting papers into structured reading notes; running results-analysis to produce strict statistics and scientific figures; generating post-experiment reports with results-report; drafting papers using skills like nature-writing; and creating evidence-anchored rebuttals via review-response. It also manages an Obsidian-based project knowledge base with /kb-* commands for on-disk note management.
- A graduate student in CS wants to start a new research topic and needs a literature-grounded plan and open questions.
- An AI researcher wants to analyze finished experiment results, compare runs, and get a decision-oriented summary.
- A researcher needs to review a Zotero collection on a topic, grouping papers and identifying gaps.
- An author wants to draft the related work section of a paper based on existing notes and findings.
- A scientist receiving reviewer comments wants to generate a structured rebuttal draft.
What are this agent's strengths and limitations?
- Covers the full research lifecycle from ideation to publication with dedicated skills and commands.
- Supports multiple CLI backends (Claude Code, Codex CLI, Kimi Code CLI, OpenCode) with separate branches.
- Deep integration with Zotero and Obsidian for literature management and durable knowledge base.
- Evidence-gated workflow ensures claims are traceable to verifiable evidence.
- Installer supports incremental updates and safe uninstall, reducing maintenance overhead.
- Primarily tailored for CS and AI research; other disciplines may require adjustments.
- Requires external tools (Zotero, Obsidian, CLI backends) increasing setup complexity.
- Not fully autonomous; requires active human decision-making.
- Need to manually merge existing local configuration files after installation.
- Documentation is extensive but may be overwhelming for new users.
How do you install or deploy this agent?
Full installation (recommended):
git clone https://github.com/Galaxy-Dawn/claude-scholar.git /tmp/claude-scholar
bash /tmp/claude-scholar/scripts/setup.shOn Windows, use Git Bash or WSL. The installer is backup-aware, preserves existing ~/.claude/CLAUDE.md, and installs the repo version as CLAUDE.scholar.md. Alternatives include minimal, selective, or plugin marketplace installation (/plugin marketplace add Galaxy-Dawn/claude-scholar).
How do you use this agent?
After installation, start by describing your task in natural language. For example:
- Start new research: "Help me start research on [topic]."
- Review Zotero collection: "Review my Zotero collection on brain foundation models."
- Analyze experiments: "Analyze the results in this experiment folder."
- Draft a paper or rebuttal: "Help me draft the related work section." Specific commands like /research-init, /zotero-review, /analyze-results are available and documented.