Claude Agent SDK
Build autonomous Claude-powered agents that inspect information, use tools and complete multi-step tasks in a managed process.
Official product pageAvery Rulebook for Claude agents
Give Claude agents an enterprise decision boundary before they act.
The Claude Agent SDK provides an agent loop with tools, files, commands, sessions and permission controls. Rulebook complements it with confirmed enterprise policy that is independent of an agent's prompt, model context and runtime process.
The operating reality
Claude agents can explore context and perform complex multi-step work. Enterprise authorization often depends on formal policy, contract terms, effective dates and named human authority that should not be inferred from the working context alone.
Keepthe product you chose
Addexecutable enterprise policy
Provewhy every action was allowed
The runtime compliance gate
Check the answer or action before it creates impact.
Rulebook evaluates live context against the confirmed policy and regulatory rules that apply, then returns a decision the agent must follow.
01Context
Actor, purpose, data, jurisdiction and proposed action
02Gate
Allow, deny, obligations or named human review
03Audit
Evidence, rulebook version and signed decision trace
Architecture boundary
Avery Rulebook
Return deterministic or evidence-backed business decisions before protected tools, conclusions and state changes.
The Claude Agent SDK owns reasoning, tool use and session execution. Rulebook owns the published enterprise rule and verifiable decision record.
Why the gap persists
Policies exist. Runtime compliance gates do not.
01
Agent instructions are not a policy approval artifact.
02
Tool permissions can require transaction-specific business context.
03
Long-running agent sessions may outlive a policy version change.
Rulebook in the workflow
Plan
The Claude agent decomposes the task and assembles relevant facts.
Authorize
A tool hook or MCP call sends the proposed action and context to Rulebook.
Execute or escalate
The agent acts only under the returned obligations, or pauses for the specified human authority.
Potential outcomes
Make compliance a gate in the work, not a report about the work.
- A stable policy boundary around autonomous tool use
- Less duplication of enterprise rules in agent instructions
- Receipts that can be correlated with Claude session and tool events
Questions leaders ask
Does Rulebook replace Claude permission controls?
No. Runtime permissions constrain what the process can access or execute. Rulebook determines whether the enterprise permits the business action in context.
Can Claude agents call Rulebook through MCP?
Yes. MCP is a natural integration surface, and REST or an SDK can be used when the host application needs tighter control.
Executive briefing
Add an auditable compliance gate to Claude Agent SDK integration.
Bring one consequential agent workflow. We will map the governing policies and regulations, runtime gate, human authority and auditable evidence path with your team.
Request a briefing