Claude Cursor Skill

researching-web

Web research via platform web tools. Use for technical comparisons, current-state

LLM Mart · 0 points · 0 0 0

Virus-scanned Reviewed automatically before listing.

Full trust report

Download alexei-led-cc-thingz-src_skills_researching-web-ce56bb4.zip · 2 KB

alexei-led/cc-thingz 36 5 forks MIT Updated 1d ago

Part of alexei-led/cc-thingz — 91 skills

Install

skills CLI npx skills add https://github.com/alexei-led/cc-thingz/tree/master/src/skills/researching-web

Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install alexei-led-cc-thingz@llmmart

Git git clone https://github.com/alexei-led/cc-thingz.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole alexei-led/cc-thingz collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Web Research

Answer with current, cited evidence and keep sourced facts apart from your recommendation. Research questions get web evidence, not answers from memory.

Never put private code, secrets, credentials, or proprietary data into a web query; answer those from local context and say what that limits.

Tools

Match the tool to the question with whatever web tools the runtime provides:

Simple fact: one focused search or answer query.
Source selection: search, then fetch the best primary sources.
Broad investigation: a deep or asynchronous research tool when available; otherwise chain searches with targeted fetches and note the fallback.
Detail on a source already found: fetch that URL instead of searching again.

Fetch enough sources to support each claim; fetch every citation only when the risk or ambiguity justifies it. For source ranking, stale-source signals, and Pi web tools, read references/sources.md.

Judgment

Prefer specifications, official docs, maintainer release notes, and primary announcements. Treat blogs and forum answers as supporting evidence.
Check sourced facts against the local project's constraints before recommending a change.
When sources conflict, cite both sides and lower confidence unless one is clearly more authoritative or current.
Flag stale-source risk when recency matters.
When live web access is unavailable, say so and label the answer as limited instead of presenting an uncited recommendation as fact.

Platform additions

No target-specific additions.

Output

## Research Result

### Research Question

<question and the decision it informs>

### Answer

<concise answer>

### Evidence

- <source title/url> — <what it supports>

### Recommendation

<recommendation separated from sourced facts, or none>

### Fit For This Repo

<what changes because of local constraints>

### Unknowns and Gaps

<conflicts, stale-source risk, missing evidence, or blocked retrieval>

Files (cc-thingz)

.agentbundler

targets

claude.json 964 B

{
  "bodyPatch": {
    "mode": "sections",
    "sections": [
      {
        "headingPath": [
          "Web Research",
          "Platform additions"
        ],
        "body": "\nThe research question is `$ARGUMENTS`.\n\nPrefer tools in this order when they are available:\n\n1. `mcp__perplexity-ask__perplexity_ask` for broad questions, comparisons, and\n   source discovery.\n2. `WebSearch` for source discovery when Perplexity is unavailable or too\n   coarse.\n3. `WebFetch` to read the primary sources you cite.\n\nUse `Task` only for broad background research that can run independently; ask\nit for cited facts, not a final recommendation.\n\n"
      }
    ]
  },
  "frontmatterPatch": {
    "allowed-tools": [
      "Task",
      "Read",
      "Grep",
      "Glob",
      "WebSearch",
      "WebFetch",
      "mcp__perplexity-ask__perplexity_ask"
    ],
    "argument-hint": "<research question>",
    "context": "fork",
    "user-invocable": true
  }
}

references

sources.md 1.3 KB

# Sources and Web Tools

## Source ranking

When several sources answer the same claim, cite the highest available:

1. Specification or standards body (RFC, W3C, ECMA, ISO).
2. Official project documentation.
3. Maintainer release notes, changelogs, or blog posts.
4. Peer-reviewed paper or authoritative book.
5. Third-party guide with an explicit version and date.
6. Recent, well-voted Stack Overflow answer.
7. Other blogs, tutorials, and forum threads.

Do not cite a rank 6–7 source as definitive when a rank 1–2 source exists.

## Stale-source signals

Flag stale-source risk when:

- The source is more than 18 months old in a fast-moving ecosystem (JS, Python
  packaging, cloud APIs, LLM tooling).
- It covers a version older than the project's current dependency.
- It predates the time horizon of the question.
- The official docs deprecate an endpoint, flag, or key it relies on.

## Pi web tools

- `web_search`: multi-query search with titles, URLs, and snippets. Use
  `search_recency_filter` when freshness matters and `search_domain_filter` to
  restrict to official docs.
- `web_answer`: grounded single-turn answer. Use it for quick facts; it does
  not replace reading primary sources.
- `web_research`: asynchronous deep investigation that delivers a report. Use
  it for open-ended or multi-step questions.

SKILL.md 2.4 KB

---
description: Web research via platform web tools. Use for technical comparisons, current-state
  and release-behavior questions, recent facts, ecosystem news, best practices, standards,
  or questions needing grounded web evidence. NOT for exact API syntax, config keys,
  or code examples — use looking-up-docs for those. NOT for repo-specific questions
  — search local files first.
name: researching-web
---

# Web Research

Answer with current, cited evidence and keep sourced facts apart from your
recommendation. Research questions get web evidence, not answers from memory.

Never put private code, secrets, credentials, or proprietary data into a web
query; answer those from local context and say what that limits.

## Tools

Match the tool to the question with whatever web tools the runtime provides:

- Simple fact: one focused search or answer query.
- Source selection: search, then fetch the best primary sources.
- Broad investigation: a deep or asynchronous research tool when available;
  otherwise chain searches with targeted fetches and note the fallback.
- Detail on a source already found: fetch that URL instead of searching again.

Fetch enough sources to support each claim; fetch every citation only when
the risk or ambiguity justifies it. For source ranking, stale-source signals,
and Pi web tools, read [references/sources.md](references/sources.md).

## Judgment

- Prefer specifications, official docs, maintainer release notes, and primary
  announcements. Treat blogs and forum answers as supporting evidence.
- Check sourced facts against the local project's constraints before
  recommending a change.
- When sources conflict, cite both sides and lower confidence unless one is
  clearly more authoritative or current.
- Flag stale-source risk when recency matters.
- When live web access is unavailable, say so and label the answer as limited
  instead of presenting an uncited recommendation as fact.

## Platform additions

No target-specific additions.

## Output

```markdown
## Research Result

### Research Question

<question and the decision it informs>

### Answer

<concise answer>

### Evidence

- <source title/url> — <what it supports>

### Recommendation

<recommendation separated from sourced facts, or none>

### Fit For This Repo

<what changes because of local constraints>

### Unknowns and Gaps

<conflicts, stale-source risk, missing evidence, or blocked retrieval>
```