research

name	research
description	Research a topic using the memory MCP and web search. Persists findings automatically.
user-invocable	true
argument-hint	[topic or question]

Research Skill

Purpose

Investigate a topic by searching the existing knowledge store (the memory MCP) and the web, then persist findings automatically.

Argument Parsing

$ARGUMENTS is the research topic or question. If omitted, use AskUserQuestion to ask what to research.

Workflow

Step 1: Search Existing Knowledge

Launch 2 researcher agents in parallel via the Agent tool (both with run_in_background: true):

Agent A — knowledge search:

• Search the memory MCP via recall over stored knowledge (documents/chunks): relationships, facts, and overviews
• Use top_k=15 for broad exploration
• Topic: $ARGUMENTS

Agent B — memory search:

• Search the memory MCP via recall for any user-related context on the topic
• Topic: $ARGUMENTS

Step 2: Gap Analysis and Contradiction Detection

When both agents return, synthesize their results:

1. List what is already known (from the memory MCP)
2. Contradiction check: if the two searches return conflicting information, flag the contradiction explicitly and investigate which is current
3. Identify gaps: what questions remain unanswered?
4. If no gaps exist, skip to Step 4

Step 3: Web Research

Launch a researcher agent to fill identified gaps:

• Provide the specific gaps as search targets
• Agent uses WebSearch to find sources, WebFetch to extract details
• Source credibility: tag each finding with source type (official docs / engineering blog / forum post / vendor marketing). Flag findings that rely on a single low-credibility source
• Agent saves new findings to the memory MCP (remember / ingest) before returning

Step 4: Report

Present findings in BLUF format:

## Conclusion
[1-2 sentence answer to the research question]

## Key Findings
[Numbered list of supporting facts/evidence]

## Sources
[Where each finding came from: the memory MCP or web URL]

## Gaps
[What remains unknown, if anything]

Step 5: Persist

If Step 3 was skipped (existing knowledge was sufficient), save the synthesized conclusion to the memory MCP via ingest — the synthesis itself is new knowledge even if the inputs were not.