AGENTS.md · CLAUDE.md · scoped rules
Repository instructions
Durable, version-controlled guidance that tells coding agents how a repository is organized, how to verify work, and which boundaries matter.
Open field guideAgent systems field guideReviewed July 26, 2026
Every layer of the agent stack has a different job.
Instruction files are not Skills. Skills are not subagents. Tools are not memory. This living library explains where each component fits, when to use it, and how to know it is production-ready.
The operating model
Use the smallest layer that fits
Start with a well-shaped task and concise project truth. Add reusable capability, delegation, external access, persistence, and automation only when the workflow earns them. Every added layer creates a new quality and security boundary.
- 01
Always-on context
Repository instructions carry concise, durable project truth.
- 02
On-demand capability
Skills and plugins load specialized procedures and resources.
- 03
Delegated execution
Subagents isolate independent work and return bounded results.
- 04
Connected, controlled action
Tools, protocols, hooks, permissions, and evals make work useful and accountable.
12
maintained field guides
24
unique primary sources
30 days
maximum review interval
Decision library
Choose the component you need to understand
Each guide includes a decision boundary, production checklist, artifact patterns, adjacent layers, and an evidence ledger.
Context
What the agent knows now and can recover later.
Working context · durable memory · retrieval
Memory and context
The systems that decide what an agent sees now, what it can retrieve later, and which durable facts persist across sessions.
Open field guideCapability
Reusable knowledge and procedures loaded when relevant.
SKILL.md · scripts · references · assets
Agent Skills
Discoverable, reusable packages of instructions and resources that teach an agent a specialized workflow without loading the whole procedure into every session.
Open field guideDistribution
How extensions are packaged, installed, and updated.
Skills · agents · hooks · connectors
Plugins
Installable bundles that distribute one or more agent extensions—such as Skills, subagents, hooks, MCP servers, or application connectors—as a coherent product.
Open field guideOrchestration
How work is delegated, isolated, and synthesized.
Delegation · isolation · parallel work
Subagents and agent teams
Specialized workers that perform bounded tasks in separate contexts and return results to a coordinating agent or shared team.
Open field guideConnectivity
How agents reach tools, data, and other agents.
Capabilities · resources · prompts · servers
Tools and MCP
Typed tools connect an agent to actions and data. The Model Context Protocol standardizes how compatible clients discover and invoke those capabilities.
Open field guideA2A · agent cards · tasks · messages
Agent-to-agent protocols
Protocols for discovering remote agents, negotiating supported interfaces, exchanging messages, and tracking long-running tasks across organizational boundaries.
Open field guideControl
Deterministic boundaries around probabilistic behavior.
Lifecycle events · policy checks · formatters
Hooks
Deterministic handlers that run at defined points in an agent lifecycle to validate, block, transform, log, or trigger work.
Open field guideLeast privilege · approvals · sandboxing · audit
Permissions and safety
The trust boundaries that constrain what an agent can read, execute, change, publish, purchase, or send on a user's behalf.
Open field guideOperations
How stable agent workflows run without constant supervision.
Schedules · triggers · unattended runs
Automations
Repeatable agent runs started by time or events, with explicit inputs, permissions, outputs, review gates, and recovery behavior.
Open field guideQuality
How teams measure, compare, and prevent regressions.
Representative tasks · graders · traces · regression gates
Evals and verification
A repeatable evidence loop that measures whether an agent completes representative tasks correctly, safely, and efficiently.
Open field guideTransactions
How agents participate in purchases with accountable consent.
UCP · AP2 · consent · mandates
Agent commerce and payments
Standards and controls for product discovery, checkout, payment authorization, and evidence of user intent in agent-mediated transactions.
Open field guideLiving documentation
Regular publishing with quality gates
AgentAtlas checks the linked primary sources every week, refuses stale catalog entries, validates citations and structure, and publishes a rotating Agent Stack Brief. Human pages, JSON APIs, feeds, sitemaps, and AI-readable indexes all use this same reviewed dataset.
Primary sources30-day review gateWeekly generated brief
Still deciding?
Compare the layers teams most often confuse
Get direct answers for AGENTS.md vs Skills, MCP vs A2A, hooks vs instructions, memory vs repository context, and more.