jina-web-research

name	jina-web-research
description	Jina Reader-first collection workflow for public web research: s.jina.ai search, Reader page reading, and X/Twitter evidence rules. Use when Jina search or X/Twitter signal is explicitly requested, or as the collection procedure referenced by the web-research skill. For general 'research this topic' requests, prefer web-research as the entry point.

Jina Web Research

Overview

Use this skill as the collection workflow for public web research that should prefer Jina Reader. It uses Jina search URLs as the primary collection path and treats X/Twitter as an indexed public-source signal, not complete platform coverage. The web-research skill is the general entry point that plans, delegates, and synthesizes; this skill defines how each search-and-read pass is actually performed, both in the main session and inside delegated subagents.

Scope

• Use public sources by default.
• Prefer Jina Reader for search and page reading.
• Use X/Twitter search when the user asks for social signal, user complaints, launch reactions, product sentiment, or current chatter.
• Cross-check important claims with non-X sources when possible.
• Do not use this for authenticated browsing, form submission, paid APIs, or internal source research unless another connector is explicitly available and allowed.

Workflow

1. Restate the research question, audience, time horizon, and scope.
2. Create a temporary research folder under tmp/research_<topic>/ when saving intermediate files.
3. Break the question into 1-5 non-overlapping subtopics.
4. Search each subtopic with Jina Reader search URLs.
5. Read promising result URLs with Jina Reader when deeper source detail is needed.
6. Cluster findings by source type and evidence strength.
7. Synthesize a cited answer that separates observed evidence from inference.

Jina Search

Prefer the Jina MCP server's direct search tool (e.g. search_web) when it is exposed. Otherwise use the Jina read_url tool with Jina search URLs, URL-encoding the full query after q=:

• General search: https://s.jina.ai/?q=<url-encoded query>
• X search: https://s.jina.ai/?q=site%3Ax.com%20<url-encoded query terms>
• Twitter fallback: https://s.jina.ai/?q=site%3Atwitter.com%20<url-encoded query terms>
• Source-specific search: add site:<domain> to the query.

X/Twitter Handling

• Always search both site:x.com and site:twitter.com when X signal matters.
• Record search terms and date searched.
• Treat indexed X results as partial evidence because login walls, deletion, ranking, and indexing gaps can hide relevant posts.
• Treat deleted, protected, suspended, login-blocked, or incomplete posts as unavailable and say so.
• Do not present search snippets as full post contents.
• Do not infer identity from a handle alone when attribution matters.
• Do not infer frequency from a few loud posts. Mark these as anecdotal unless corroborated.
• Prefer direct post URLs when available; otherwise cite the Jina search result URL and explain the limitation.
• Do not use a logged-in browser session just to bypass protected or deleted X content unless the user explicitly asks.

Evidence Rules

• Cite URLs for every material claim.
• Separate facts, quotes, and inference.
• Distinguish official sources, journalism/blogs, forums, GitHub, Reddit/HN, and X/Twitter.
• Call out missing or weak source access.
• Do not overclaim recency; include concrete dates when dates matter.
• Do not store API keys, cookies, bearer tokens, or other secrets in research files or repository artifacts.

Output

Default to an in-chat brief unless the user asks for a saved report.

Include:

• Executive summary.
• Findings ranked by confidence and relevance.
• Source map showing what was searched and what each source contributed.
• X/Twitter signal section when searched.
• Gaps, caveats, and recommended next checks.