What does this skill do, and when should you use it?
autoresearch is one skill in the claude-obsidian collection. It performs iterative web research for a topic, then files sources, concepts, entities, and a synthesis into an Obsidian knowledge base. Objectives, source preferences, confidence rules, round limits, and page limits are configured through program.md. It is a good fit for users who want research to compound into an owned Markdown knowledge graph, but it depends on the collection’s vault structure, scripts, and web tools.
Reads references/program.md and vault mode configuration; decomposes a topic into multiple search angles; calls WebSearch and WebFetch; runs up to three rounds of broad search, gap filling, and synthesis checking; validates URLs, sanitizes fetched content, and limits body size; routes output paths according to the active methodology; coordinates concurrent writes with file locks; creates source, concept, entity, and “Research: [topic]” synthesis pages; updates wiki/index.md, wiki/log.md, and wiki/hot.md.
- A researcher wants to turn papers or source material about a topic into cited Obsidian research pages.
- A knowledge-management user needs a multi-pass web investigation of an unfamiliar subject with automatic Markdown filing.
- A business analyst wants to preserve source conflicts, evidence gaps, and unresolved questions.
- A user of Generic, LYT, PARA, or Zettelkasten organization wants research filed according to the vault’s active method.
How do you install this skill?
- Before running, disclose the expected network-request and write volume and confirm permission to file fetched content into the vault.
- Do not provide credentials, sensitive personal data, or internal URLs to the loop; the materials do not establish sensitive-data isolation.
- In mainland-China environments, WebSearch/WebFetch and some external sources may be unreachable; prepare alternative sources or manual imports.
- Run in a branch or recoverable backup to reduce recovery cost from bulk writes and automatic commits.
- Shell / CLI
- Network access
- Local filesystem
- MCP Server
Python 3BashObsidian vaultWebSearchWebFetch
The skill is distributed with the 15-skill AgriciDaniel/claude-obsidian repository. For the public build:
- git clone https://github.com/AgriciDaniel/claude-obsidian
- cd claude-obsidian
- bash bin/setup-vault.sh
- Open the folder as an Obsidian vault and open Claude Code in the same folder.
The collection can also be installed as a plugin with claude plugin marketplace add AgriciDaniel/claude-obsidian followed by claude plugin install claude-obsidian@agricidaniel-claude-obsidian. The source does not document a standalone autoresearch installation.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/AgriciDaniel/claude-obsidian.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/autoresearch" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- deep dive into [topic]
Configure research objectives and constraints in skills/autoresearch/references/program.md. In Claude Code, enter /autoresearch [topic], or use triggers such as “research [topic]”, “deep dive into [topic]”, or “investigate [topic]”. After completion, inspect the generated pages under wiki/questions, wiki/sources, wiki/concepts, and wiki/entities, along with the updated index, log, and hot cache.
What are this skill's strengths and limitations?
- Supports up to three rounds of searching, fetching, gap filling, and synthesis.
- Separates output into source, concept, entity, and synthesis pages with cross-references.
- Includes URL validation, fetched-content sanitization, body-size limits, and failure logging.
- Supports Generic, LYT, PARA, and Zettelkasten filing modes.
- Produces plain Markdown files owned by the user.
- Depends on Claude Code-style tools, the collection’s vault structure, scripts, and configuration files rather than being a standalone skill.
- A full run can involve roughly 45 WebFetch calls, making high-cost topics expensive.
- Requires network access and may use Bash, Python 3, Obsidian CLI, or an MCP transport.
- The source provides no standalone test suite or production-verification evidence for other platforms.
- Research quality depends on source availability, program.md configuration, and search results.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Autonomous Research Filing this page | 52 · Use with care | ★ 16k | 1mo ago | MIT |
| Wiki Knowledge Companion | 49 · Use with care | ★ 16k | 1mo ago | MIT |
| Compound Vault Methodology Modes | 56 · Use with care | ★ 16k | 1mo ago | MIT |
| Wiki Log Folder | 55 · Use with care | ★ 16k | 1mo ago | MIT |
| LLM Wiki Manager | 51 · Use with care | ★ 71 | 1mo ago | MIT |
The README positions the project as an open-source local Markdown knowledge base and Notion alternative, and compares the collection with Smart Connections and Copilot on automation, contradiction flagging, session memory, and autonomous research. Those comparisons concern the full claude-obsidian collection; this skill itself is limited to autonomous research and filing.
How did FollowSkills review this skill?
The skill specifies URL-scheme and private-network rejection, host restrictions, content sanitization, failure logging, and per-file locking. However, WebSearch/WebFetch target arbitrary external sources and the loop can perform many writes to the user's vault without explicit per-write confirmation, with no clear sensitive-data policy, complete rollback procedure, or credential isolation; points are deducted for those gaps.
Rounds, source limits, routing, transport fallback, failure continuation, and diagnostic logging are specified, making the happy path reasonably coherent from a static read. The selected skill has no dedicated tests, and its routing, locking, transport, and filing dependencies are not reproduced here; abnormal-input and tool-failure coverage remains incomplete, so the static ceiling applies.
The intended use—multi-round web research for a topic with structured Obsidian filing—is clear, with triggers, input paths, and several organization modes documented. Trigger phrases are broad and non-fit boundaries are limited; core operation depends on web search and fetching whose reachability from mainland-China networks is not established, so points are deducted.
The skill has progressive sections, a configurable program, output templates, constraints, failure behavior, and version references. Repository context also supplies an MIT license, security policy, CI, and README documentation. The selected skill does not provide its own changelog, explicit maintenance owner, or update path, and publisher identity is unverified by FollowSkills, so full marks are not justified.
The defined outputs—source, concept, entity, synthesis, index, log, and hot-cache updates—are directly usable in the stated wiki workflow, and the program defines source preferences and confidence rules. Static review cannot verify representative results, and the high fetch/write cost may be disproportionate for some topics; the static maximum is therefore used.
The repository includes CI, static validation, several script tests, and a pinned revision. Those tests do not cover autoresearch's search, fetch, sanitization, routing, filing, or synthesis paths, and no third-party execution evidence is provided, so the score remains limited.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →