mmx-text-researcher

name

mmx-text-researcher

description

Deep research skill using MiniMax CLI (mmx). Activates when: (1) Arif asks to research, investigate, look up, find information, or deep-dive on a topic; (2) a question requires synthesis from multiple sources; (3) Arif needs structured output (JSON, tables, summaries); (4) any knowledge task that benefits from web search + synthesis. Uses mmx search + mmx text chat. NOT for simple factual lookups (use web_search tool directly).

MMX Text Researcher — Deep Research via MiniMax CLI

Use this skill when a question requires investigation, synthesis, multi-source reasoning, or structured output.

Prerequisites

• mmx CLI must be authenticated: mmx auth status
• If not authenticated: mmx auth login --api-key <key>
• Check region: mmx config show (global vs CN)

Core Research Loop

1. SEARCH  → mmx search "<query>" [--output json]
2. SYNTH   → mmx text chat --model MiniMax-Text-01 --message "Summarize from: <results>"
3. STRUCT  → mmx text chat --output json --message "Extract key facts as JSON"
4. VERIFY  → mmx search "<specific claim>" to cross-check
5. COMPILE → Final synthesis in requested format

Commands Reference

Web Search (structured JSON output)

mmx search "MiniMax AI latest developments"
mmx search query --q "specific fact to verify" --output json

Deep Research Chat (MiniMax-Text-01 best for synthesis)

mmx text chat --model MiniMax-Text-01 --message "Research brief: <topic>. Provide structured output with: key findings, confidence level per finding, sources, and remaining unknowns."

Streaming Research (for long investigations)

mmx text chat --model MiniMax-Text-01 --message "<research query>" --stream

Multi-turn Research Thread

mmx text chat --message "I want to research X. Start by identifying the 5 most important sub-questions."
mmx text chat --message "Now investigate sub-question 1: <question>"
# ... continue threading

Structured Output (JSON)

mmx text chat --model MiniMax-Text-01 --output json --message "Extract into JSON: { topic, findings: [{ claim, confidence, source }], gaps: [], next_steps: [] }"

Fact Verification Loop

For claims made in previous research:

mmx search "specific claim to verify"
mmx text chat --message "Based on search results: does this support or refute the claim: <claim>? Cite sources."

Output Format Options

Specify format in the prompt:

• JSON — structured machine-readable output
• table — tabular comparison or summary
• brief — executive summary (3-5 bullet points)
• full — comprehensive research report
• verdict — true/false/uncertain with reasoning

Quality Checklist

• At least 2 independent sources per key claim
• Confidence level assigned per finding
• Unknowns/gaps explicitly stated
• Sources cited (URL or source name)
• Opposing views included if relevant

Common Triggers

• "research X for me"
• "investigate this claim"
• "find information about"
• "deep dive into"
• "what is the current state of"
• "synthesize what we know about"
• "verify whether"
• "compare X vs Y"