Agents — Taxonomy, Communication, and Identity
The section that addresses "who, what, on whose behalf, and how they coordinate." Deepens the Agent layer of the three-layer model (03-architecture).
Position of This Section
The Agents section is being grown incrementally toward the Agent ID era. It is organized: terminology → implementation units → communication protocols → identity.
Page Overview
Concept layer — Get terminology straight first
| Page | Theme | Best for |
|---|---|---|
| Agent Taxonomy | Organize custom/sub/meta-agents, Orchestrator-Worker, Swarm, etc., into 4 abstraction levels | Anyone wanting cross-framework terminology |
Implementation layer — What to build and how
| Page | Theme | Best for |
|---|---|---|
| Custom Sub-agent | Sub-agent definition, placement, and use in Claude Code | Foundational understanding before implementation |
| Sub-agent vs Skills | Whether to implement as a Skill or a Sub-agent; composition patterns | When you're stuck on a design decision |
| Quality Gate Pattern | Validator sub-agent implementation, CI/CD integration, pass criteria | When self-review keeps coming out too lenient |
| Multi-Agent / Agent Teams | Beyond single sub-agents — Orchestrator-Worker, Hierarchical, Swarm patterns | When single agents hit a ceiling |
Communication layer — How agents talk to each other
| Page | Theme | Best for |
|---|---|---|
| What is A2A (Agent-to-Agent Protocol) | A2A v1.0 overview, standardization under Linux Foundation, complementarity with MCP | Anyone exploring cross-org agent coordination |
Identity layer — On whose behalf do they act?
| Page | Theme | Best for |
|---|---|---|
| Agent Identity | Non-Human Identity as a new category, OpenID Foundation's 4 architectural philosophies, commercial implementation status | Operators bringing agents to production |
Roadmap Toward the Agent ID Era
This site grows alongside the 2025–2026 Agent ID standardization wave. Here is the honest current status and what's coming.
| Topic | Status |
|---|---|
| Agent Taxonomy | ✅ Published |
| Custom Sub-agent (Claude Code) | ✅ Published |
| Sub-agent vs Skills | ✅ Published |
| Quality Gate Pattern | ✅ Published |
| Multi-Agent / Agent Teams | ✅ Published |
| A2A Protocol | ✅ Published |
| Agent Identity (identification & delegation) | ✅ Published |
| Permission management (RBAC / ABAC / JIT) | 🚧 Planned — referenced from agent-identity |
| Agent governance | 🚧 Planned |
| A2A implementation patterns (Web Bot Auth, Macaroons, etc.) | 🚧 Planned |
| Agent Marketplace / Registry | 🚧 Under consideration |
IMPORTANT
The Agent ID area has rapidly moved into production-operation phase: OpenID Foundation white paper v1.1 (October 2025), Microsoft Entra Agent ID GA (April 2026), Okta for AI Agents GA, A2A v1.0 under Linux Foundation. Because specs remain fluid, this site distinguishes "what is settled now" from "what remains open."
Recommended Reading Routes
Route 1: Newcomer (foundation)
Route 2: Production readiness (identity-first)
Route 3: At scale (multi-agent coordination)
Route 4: Raising quality (implementation techniques)
Related Sections
| Section | Relation |
|---|---|
| Concepts / 03-architecture | The whole three-layer model including the Agent layer |
| Concepts / 07-doctrine-and-intent | Designing "constraints and objectives" given to agents |
| Concepts / 08-memory-and-knowledge | The memory layer agents reference |
| Skills | Static knowledge agents reference |
| MCP | How agents connect to the outside world |
| FAQ / Agent vs Sub-agent vs Skill vs MCP | The 3-line answer for all four |
🔗 Deeper: From the context-management perspective
For agents' essential constraints — why isolated contexts matter, why Multi-Session coordination is needed — grounded in LLM structure, see the sister site.
- understanding-llm / Part 5: On-Demand Context (Skills & Agents) — Agents that expand only when requested
- understanding-llm / Part 10: Multi-Session Coordination (Agent Teams) — Scaling beyond a single agent