Skip to content

strands.vended_interventions.cedar.cedar_authorization

Cedar authorization intervention handler.

TypeAndId

Section titled “TypeAndId”

A Cedar entity identifier with ‘type’ and ‘id’ keys.

PrincipalResolver

Section titled “PrincipalResolver”

Resolves a principal from invocation_state. Return None to deny (fail-closed).

ContextEnricher

Section titled “ContextEnricher”

Injects extra fields into context.session. Receives {‘tool_name’, ‘tool_input’, ‘invocation_state’}.

CedarAuthorization

Section titled “CedarAuthorization”
class CedarAuthorization(InterventionHandler)

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:71

Cedar authorization intervention handler.

Evaluates Cedar policies before each tool call. Each tool maps to a Cedar action, with context structured as \{input: <tool_args>, session: \{hour_utc, call_count, ...}}.

Call counts are persisted to agent.state under the key "cedar-authorization" so they survive handler recreation and are included in session snapshots.

.. note:: Each handler instance is scoped to a single agent. Sharing one instance across multiple agents will cause rate-limit counts to leak between them.

Example::

cedar = CedarAuthorization( policies=‘permit(principal, action == Action::“search”, resource);’ ) agent = Agent(interventions=[cedar], tools=[search_tool])

on_error

Section titled “on_error”
@property
def on_error() -> OnError

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:95

What to do when this handler throws.

__init__

Section titled “__init__”
def __init__(*,
             policies: str,
             tools: list[ToolDefinition] | None = None,
             entities: list[dict[str, Any]] | str | None = None,
             schema: str | None = None,
             principal: TypeAndId | None = None,
             principal_resolver: PrincipalResolver | None = None,
             context_enricher: ContextEnricher | None = None,
             on_error: OnError = "throw") -> None

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:99

Initialize the Cedar authorization handler.

Arguments:

  • policies - Inline Cedar policy text or path to a .cedar file.

  • tools - MCP tool definitions for auto schema generation.

  • entities - Entity data as inline list, JSON string, or path to .json file.

  • schema - Cedar schema as inline text or path to .cedarschema file.

  • principal - Static principal identity.

  • principal_resolver - Dynamic principal resolver from invocation_state.

  • context_enricher - Callback to inject extra fields into context.session.

  • on_error - What to do when principal_resolver or context_enricher raises: 'throw' (default) re-raises, 'deny' denies the tool call (fail-closed, recommended alternative to the default).

    Warning: 'proceed' is fail-open. When a callback raises, the tool call runs without any Cedar authorization check. Never use it when the agent relies on Cedar to enforce access.

before_tool_call

Section titled “before_tool_call”
def before_tool_call(event: BeforeToolCallEvent,
                     **kwargs: Any) -> Proceed | Deny

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:161

Evaluate Cedar policy before tool execution.

reload

Section titled “reload”
def reload() -> None

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:236

Reload policies/entities/schema from disk. Validates before committing.

reset_call_counts

Section titled “reset_call_counts”
def reset_call_counts(agent: Agent | None = None) -> None

Defined in: src/strands/vended_interventions/cedar/cedar_authorization.py:256

Clear all rate-limit call counters.

Arguments:

  • agent - If provided, also clears persisted counts from agent.state.