Skip to content

RECON

research

Deep technical research that auto-saves to a knowledge base. Never research the same thing twice.

RECON conducts comprehensive research using Claude's built-in research agent, then saves the results persistently. Next time the topic comes up — in this session or any future session — the knowledge is already there.

Usage

/recon WebSocket authentication best practices

/recon --quick difference between JWT and session tokens

/recon --deep Rust async runtime comparison: tokio vs async-std vs smol

/recon --compare GraphQL vs REST for mobile APIs

What Happens

  1. Check existing knowledge — is this already researched?
  2. Conduct research — launches Claude's technical-research-agent
  3. Generate report — structured markdown with findings
  4. Save to knowledge base — knowledge/research/YYYY-MM-DD_topic-slug.md
  5. Update index — knowledge/index.json for future searchability

Report Structure

Every RECON report includes:

Section Contents
Summary TL;DR of findings
Key Findings Bullet points of the most important discoveries
Deep Dive Detailed analysis with code examples
Comparison Side-by-side evaluation (if applicable)
Recommendations What to actually do
Sources Where the information came from

Pairs With RECALL

RECON saves knowledge. RECALL retrieves it.

/recon WebSocket authentication       # Research and save
# ... months later ...
/recall WebSocket auth                 # Find it instantly

See the Deep Research Pipeline recipe for the full workflow.

Prerequisites

  • Claude Code with Task tool (built-in technical-research-agent)
  • Web search access
  • knowledge/ directory auto-created on first use

Research once. Know forever.