web-research
Skill103 starsChanged 7 days agoMITView on GitHub
Evidence-driven web and open-source research for questions that require multiple sources, factual verification, comparisons, or an adopt/adapt/build decision. Use for substantial research in foreground sessions, the genesis-researcher subagent, and research-profile background sessions. Skip for a single stable fact or a known URL that only needs fetching.
45 other projects use a version of this.
Adopted by
Most recently taken up by
- juninmd/skillstheir own version, first seen Nov 15, 2026
- oleg494/coding-kittheir own version, first seen Nov 8, 2026
- thensls/nsls-builder-toolkittheir own version, first seen Nov 8, 2026
- memi-design/design-skillstheir own version, first seen Nov 8, 2026
- ai-assist-org/ai-assisttheir own version, first seen Nov 8, 2026
- Bobagi/claude-skillstheir own version, first seen Nov 1, 2026
Used with
Projects with it also use data-analysis, meeting-notes and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
shanggqmSkilldata-analysis
数据分析技能,提供结构化的数据查询、统计分析和报告生成工作流。当用户需要查询数据库、进行数据统计分析、生成分析报告时使用。按规范输出原始数据、计算结果、Markdown报告和HTML可视化报告。
Also in 5 projects that use this
claude-office-skillsSkillmeeting-notes
>
Also in 4 projects that use this
MicrosoftSkillweb-search
Also in 5 projects that use this
kongyo2Skillprompt-engineering
Comprehensive prompt engineering framework for designing, optimizing, and iterating LLM prompts. This skill should be used when users request prompt creation, optimization, or improvement for any LLM task, or when users need help translating vague requirements into effective prompts through collaborative dialogue and iterative refinement.
Also in 4 projects that use this
rogue-dev-studioSkillbrowser-automation
Canonical browser automation and web UI verification: Playwright, agent browser CLIs, and CDP-based tooling in one playbook.
Also in 3 projects that use this
JetBrainsSkillbrand-guidelines
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Also in 4 projects that use this
Other versions45 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 45
- langchain-ai/deepagents★ 31ksame name
- pollinations/pollinations★ 5.3ksame name
- langchain-ai/deepagentsjs★ 1.6ksame name
- alfredxw/denova★ 1ksame name
- icey1287/SuperMew★ 884same name
- cohen-liel/hivemind★ 108same name
- grandamenium/cortextos★ 106same name
- jezweb/vite-flare-starter★ 49same name
- dewil/ai-toolkit★ 9same name
- dewil/claude-toolkit★ 9same name
35 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- genesis-external-clientSkill
- build-sessionSkill
- cc-updateSkill
- closing-sessionSkill
- code-intelligenceSkill
- content-publishSkill
- deliverable-builderSkill
- genesis-developmentSkill
- genesis-voiceSkill
- inbox-reviewSkill
- shelveSkill
- spec-sessionSkill
- subsystem-mapSkill
- tasteSkill
- unshelveSkill
- validating-mergesSkill
- voice-masterSkill
- youtube-fetchSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 38 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 21
Search with web_search, then fetch the sources that carry the actual claim with web_fetch.
- Distinctive: line 23
Prefer primary sources for product behavior, specifications, code, pricing, and current policy.
- Distinctive: line 30
Treat a failed search or tool call as unknown.
“Common” is measured: kept by at least 6 in 10 of the 38 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Web Research
Frame the decision
State the question, the decision it informs, the required freshness, and what would count as enough evidence. Choose the lightest depth that can answer it:
- Focused: one narrow claim or known source. Verify directly and stop.
- Substantial: several claims, unfamiliar subject matter, comparison, or recommendation. Decompose into independent questions and synthesize them.
- Exhaustive: high-consequence or broad landscape work. Add coverage tracking, competing explanations, and a final gap pass.
Depth follows the task. Do not add an extra pass merely because the word “research” appears.
Gather evidence
- Decompose substantial work into questions whose results can be checked independently. Parallelize only independent searches; follow-ups that depend on an earlier result stay sequential.
- Search with
web_search, then fetch the sources that carry the actual claim withweb_fetch. Use browser tools only when interaction or rendered state is necessary. - For software capability questions, search for existing implementations before proposing a build. Use
recon_github_searchfor public repositories and issues, thenrecon_github_readto inspect repository metadata, trees, and source. Repository search accepts GitHub query syntax; issue search treatsqueryas literal text and exposes repository, state, and labels as structured filters. GitHub's code-search API requires credentials, so public-only research must discover candidate repositories first and inspect their trees/files rather than silently inheriting operator access. When the capability could run inside a Claude Code session (a hook or guard, status line, prompt or context injection, tool interception, notifications, an in-terminal pane), also search Claude Code mods and plugins: themod-finderplugin'ssearch_modstool if installed, otherwise the catalogue athttps://raw.githubusercontent.com/karanb192/awesome-claude-code-mods/main/catalogue.md, plus Anthropic's directory (https://raw.githubusercontent.com/anthropics/claude-plugins-official/main/.claude-plugin/marketplace.json, reachable withweb_fetchfrom any session). Record each candidate's access footprint and the CC version it validates on; mods run unsandboxed. A background session cannot search the community catalogue yet (#3321): report that gap rather than "no mod exists". When the foreground/evaluatecommand is available, apply it for the final adopt/adapt/build comparison. In background sessions, perform that comparison directly: assess how each candidate helps, does not help, could help through adaptation, and what Genesis can learn from it; then classify it as adopt, adapt, learn from, or build. - Prefer primary sources for product behavior, specifications, code, pricing, and current policy. Use strong secondary sources for context or independent testing. A primary source can establish its own documented behavior; independent corroboration matters when incentives, interpretation, or real-world performance are disputed.
- Record the URL, publication or update date when relevant, the exact claim supported, and limits or conflicts. A search snippet is a lead, not evidence.
Evidence standard
- Separate observed facts, source claims, and inference.
- Resolve conflicts by checking recency, authority, methodology, and scope. Report material conflicts that remain.
- Treat a failed search or tool call as unknown. Never turn it into “nothing exists.”
- For recommendations, compare candidates against the stated requirements, including maintenance, license, integration cost, and reversibility.
- Calibrate confidence per major conclusion: high when evidence is direct and current, medium when evidence is indirect or incomplete, low when key uncertainty remains.
Deliver
Lead with the answer and the decision-relevant evidence. Cite every externally checkable claim near the claim. Include the searches or coverage boundaries only when they explain confidence or remaining uncertainty. End with concrete next actions and open questions when either exists.
For autonomous blocker research, preserve the caller's required JSON result shape after applying this method.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

