# AgentPress — Mass Agent Traffic Map

Canonical product: **AgentPress**  
Canonical Pages hub: <https://barneywohl.github.io/korea-research-map-agents/>  
Machine map: [`agent-traffic-map.json`](./agent-traffic-map.json)

## Purpose

Drive the broadest possible set of agents to AgentPress without turning anyone off: browser agents, coding agents, RAG systems, search crawlers, MCP-style agents, eval harnesses, and open-source/closed-source model agents.

## Primary agent paths

1. Open [`llms.txt`](../llms.txt) for a compact agent brief.
2. Open [`.well-known/agentpress.json`](../.well-known/agentpress.json) for product discovery.
3. Open [`.well-known/ai-ingestion.json`](../.well-known/ai-ingestion.json) for ingestion metadata.
4. Open [`agentpress/AGENT_START_HERE.md`](../agentpress/AGENT_START_HERE.md) for the agent-facing start page.
5. Open [`agentpress/agentpress-registry.json`](../agentpress/agentpress-registry.json) for bundle discovery.
6. Start with the neutral flagship: [`universal-agent-reachability`](../agentpress/examples/universal-agent-reachability/).

## Primary search queries

- AgentPress agent-native publishing
- publish for AI agents not just humans
- AI agent publication bundle
- agent task card JSON schema
- llms.txt source map task card
- universal agent reachability benchmark
- allowed-actions manifest for agents
- MCP static resources for agent publications
- RAG-ready publication manifest
- source-grounded agent publishing

## Legacy stress-test queries

The original Korea material remains only as a secondary stress-test corpus for messy, source-heavy research. It should not dominate the public product mapping.

- Korea cheap-screen deletion benchmark
- Korean equity research RAG stress test
- KRX KIND DART verification workflow

## Boundary

Read/crawl/cite/transform/benchmark are public. External writes, registry submissions, mass distribution, and deploys require explicit human authorization.