How do I install the Web Research skill?

Run npx skillmds@latest add ryanthedev/web-research in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Web Research safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Web Research?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Web Research free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Web Research?

ryanthedev (@ryanthedev) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Web Research by ryanthedev — Agent Skill

Name: Web Research
Author: ryanthedev

Web Research

Runs parallel multi-angle web research at four depth modes (scan, brief, breadth, deep): a planner grounds the query in local context and plans search dimensions, parallel search agents pull verbatim extracts with source URLs, one dimension always hunts counter-evidence, and synthesis is grounded back in the local project. Use when researching a topic, library, or tool on the web, comparing options or products, surveying a landscape of approaches, or checking current, version-sensitive, or time-sensitive information. Not for: driving or automating a live browser session (use oberskills:browser), academic literature search across arXiv or Semantic Scholar, or questions about the current repo, project, or files already on disk — read those directly.

ryanthedev 6079fa7 4 files · 15.6 KB Updated Sep 17, 2026

File contents

web-research

Parallel, multi-angle web search. Ground results in local context and return only the relevant findings, each one carrying a source URL.

Depth Modes

If the user specifies a depth, use it. Otherwise, infer from the query or ask:

Mode	When	Output
scan	Quick answer, sanity check, "is this a thing?"	1-pager: top 3 findings + URLs
brief	Standard research question	Synthesized brief with recommendations
breadth	Map a space, survey options, "what's out there?"	Landscape map: categories, key players, gaps
deep	Decision-critical, needs verification	Full report with confidence levels and verified sources

Inference rules:

"Quick" / "is there" / "does X support" / "what's the syntax for" → scan
"Research" / "how to" / "best way to" / no depth cue → brief
"Survey" / "landscape" / "what are the options" / "compare" → breadth
"Deep dive" / "report" / "I need to make a decision about" / "thorough" → deep

After selecting mode, load the mode's prompts:

scan/brief: references/search-prompts.md in this skill directory + references/synthesis-prompts.md in this skill directory
breadth: same two files (breadth-specific sections)
deep: same two files + references/deep-mode.md in this skill directory

Pipeline

Step 0: Check Existing Knowledge (skip for scan)

Before hitting the web, ground in what's already available.

Check local files relevant to the query (package.json, configs, code, docs).
Use what you find to skip dimensions already covered and focus the search on gaps.

If existing knowledge covers the query: Present it. Ask whether they want fresh web results.

Step 1: Plan

Dispatch planner (sonnet) with the host's subagent/delegation tool when available; otherwise plan inline. Grounds in local context, plans dimensions based on mode:

scan: 1 focused dimension
brief: 2-3 dimensions targeting different angles
breadth: 5-8 dimensions covering the landscape
deep: 3-5 high-value dimensions, prioritized by quality likelihood

Show the user the plan before dispatching search agents.

See search-prompts.md for the planner prompt template.

Step 2: Search

Dispatch search agents in parallel with the host's subagent/delegation tool when available; otherwise run the searches inline and keep artifacts small. Each writes to ~/.local/state/web-research/{timestamp}-{query-slug}-{dimension}.md

If a search returns nothing, reformulate the query and retry once. If still nothing, report the gap.

See search-prompts.md for per-mode agent prompt templates.

Step 3: Cross-Pollinate (deep mode only)

Agents review each other's findings and fill gaps. See deep-mode.md.

Step 4: Synthesize

See synthesis-prompts.md for per-mode templates. Scan mode skips this (return search results directly).

Synthesis agent inherits the user's model. Never override it.

Step 5: Verify (deep mode only)

Spot-check URLs and version numbers after synthesis. See deep-mode.md.

Handling Failures

Situation	Action
Search agent returns nothing	Reformulate and retry once. Report the gap if still nothing.
Search agent fails	Continue with others. Note the failed dimension.
All agents fail	Report failure with attempted queries. Suggest alternatives.
URL 404s during verification	Flag it. Remove from recommendations if sole source.

Model Selection

Agent	Model	Why
Planner	sonnet	Planning doesn't need the strongest model
Search agents	sonnet	Extraction quality needs reasoning depth
Cross-pollination	sonnet	Gap-filling is search, not synthesis
Synthesis	(inherit)	Never override the user's model

Search agents return file paths only, not full page content.

---name: web-researchdescription: Runs parallel multi-angle web research at four depth modes (scan, brief, breadth, deep): a planner grounds the query in local context and plans search dimensions, parallel search agents pull verbatim extracts with source URLs, one dimension always hunts counter-evidence, and synthesis is grounded back in the local project. Use when researching a topic, library, or tool on the web, comparing options or products, surveying a landscape of approaches, or checking current, version-sensitive, or time-sensitive information. Not for: driving or automating a live browser session (use oberskills:browser), academic literature search across arXiv or Semantic Scholar, or questions about the current repo, project, or files already on disk — read those directly.---# web-researchParallel, multi-angle web search. Ground results in local context and return only the relevant findings, each one carrying a source URL.---## Depth ModesIf the user specifies a depth, use it. Otherwise, infer from the query or ask:| Mode | When | Output ||------|------|--------|| **scan** | Quick answer, sanity check, "is this a thing?" | 1-pager: top 3 findings + URLs || **brief** | Standard research question | Synthesized brief with recommendations || **breadth** | Map a space, survey options, "what's out there?" | Landscape map: categories, key players, gaps || **deep** | Decision-critical, needs verification | Full report with confidence levels and verified sources |**Inference rules:**- "Quick" / "is there" / "does X support" / "what's the syntax for" → **scan**- "Research" / "how to" / "best way to" / no depth cue → **brief**- "Survey" / "landscape" / "what are the options" / "compare" → **breadth**- "Deep dive" / "report" / "I need to make a decision about" / "thorough" → **deep****After selecting mode, load the mode's prompts:**- scan/brief: `references/search-prompts.md` in this skill directory + `references/synthesis-prompts.md` in this skill directory- breadth: same two files (breadth-specific sections)- deep: same two files + `references/deep-mode.md` in this skill directory---## Pipeline### Step 0: Check Existing Knowledge (skip for scan)Before hitting the web, ground in what's already available.1. Check local files relevant to the query (package.json, configs, code, docs).2. Use what you find to skip dimensions already covered and focus the search on gaps.**If existing knowledge covers the query:** Present it. Ask whether they want fresh web results.### Step 1: PlanDispatch planner (sonnet) with the host's subagent/delegation tool when available; otherwise plan inline. Grounds in local context, plans dimensions based on mode:- scan: 1 focused dimension- brief: 2-3 dimensions targeting different angles- breadth: 5-8 dimensions covering the landscape- deep: 3-5 high-value dimensions, prioritized by quality likelihoodShow the user the plan before dispatching search agents.See search-prompts.md for the planner prompt template.### Step 2: SearchDispatch search agents in parallel with the host's subagent/delegation tool when available; otherwise run the searches inline and keep artifacts small. Each writes to `~/.local/state/web-research/{timestamp}-{query-slug}-{dimension}.md`If a search returns nothing, reformulate the query and retry once. If still nothing, report the gap.See search-prompts.md for per-mode agent prompt templates.### Step 3: Cross-Pollinate (deep mode only)Agents review each other's findings and fill gaps. See deep-mode.md.### Step 4: SynthesizeSee synthesis-prompts.md for per-mode templates. Scan mode skips this (return search results directly).Synthesis agent inherits the user's model. Never override it.### Step 5: Verify (deep mode only)Spot-check URLs and version numbers after synthesis. See deep-mode.md.---## Handling Failures| Situation | Action ||-----------|--------|| Search agent returns nothing | Reformulate and retry once. Report the gap if still nothing. || Search agent fails | Continue with others. Note the failed dimension. || All agents fail | Report failure with attempted queries. Suggest alternatives. || URL 404s during verification | Flag it. Remove from recommendations if sole source. |---## Model Selection| Agent | Model | Why ||-------|-------|-----|| Planner | sonnet | Planning doesn't need the strongest model || Search agents | sonnet | Extraction quality needs reasoning depth || Cross-pollination | sonnet | Gap-filling is search, not synthesis || Synthesis | (inherit) | Never override the user's model |Search agents return file paths only, not full page content.

Frequently asked questions