Event Types & Enums

These types are shared across the backend, TypeScript SDK, and Python SDK.

EventType

The type of event being sent to Sequenceware.

Value	When to send
run.started	Agent begins a task
run.completed	Agent finishes successfully
run.failed	Agent encounters an error
step.started	Agent begins a sub-task / phase
step.completed	Agent finishes a sub-task
tool.called	Agent is about to use a tool
tool.completed	Tool execution finished
tool.blocked	Tool call was blocked by a policy
risk.detected	Risk event flagged
approval.requested	Approval request created
approval.approved	Approval granted
approval.rejected	Approval denied
policy.matched	A policy matched a tool call

import { EventType } from '@sequenceware/sdk';

EventType.RUN_STARTED       // 'run.started'
EventType.RUN_COMPLETED     // 'run.completed'
EventType.RUN_FAILED        // 'run.failed'
EventType.TOOL_CALLED       // 'tool.called'
EventType.TOOL_COMPLETED    // 'tool.completed'
EventType.TOOL_BLOCKED      // 'tool.blocked'
// ...

PolicyAction

The action a policy takes when it matches a tool call.

Value	Description
allow	Let the tool call proceed
warn	Let it proceed, flag in dashboard
block	Reject the tool call immediately
require_approval	Pause until a human approves

import { PolicyAction } from '@sequenceware/sdk';

ApprovalStatus

The status of an approval request.

Value	Description
pending	Waiting for a reviewer
approved	Reviewer approved
rejected	Reviewer rejected

import { ApprovalStatus } from '@sequenceware/sdk';

ToolCallStatus

The status of a tool call after policy evaluation.

Value	Description
pending	Being processed
completed	Execution finished
blocked	Blocked by a policy
awaiting_approval	Paused, waiting for approval

import { ToolCallStatus } from '@sequenceware/sdk';

RiskSeverity

Risk severity levels.

Value	Description
low	Minor concern
medium	Moderate concern
high	Significant concern
critical	Immediate attention required

import { RiskSeverity } from '@sequenceware/sdk';

ActionType

Semantic action types for tool calls. Used with the actionHint parameter.

Value	Description
pr_create	Creating a pull request
pr_merge	Merging a pull request
pr_update	Updating a pull request
branch_delete	Deleting a branch
deploy	Deploying to an environment
script_execute	Executing a script
db_migrate	Running database migrations
config_change	Changing configuration
secret_access	Accessing secrets/credentials
api_write	Writing via an external API
unknown	Unclassified action

import { ActionType } from '@sequenceware/sdk';

AgentContext

Provides additional context about the agent's intent and reasoning. Can be attached to runs and tool calls to improve observability and audit quality.

interface AgentContext {
  goal?: string;           // User's original task/instruction
  reasoning?: string;      // Agent's chain-of-thought before this action
  conversationId?: string; // External reference to conversation thread
  promptSummary?: string;  // Summary of active prompt/context
}

Field	Type	Description
goal	string?	The user's original task or instruction the agent is working on
reasoning	string?	The agent's chain-of-thought or rationale before taking this action
conversationId	string?	External reference to the conversation thread (e.g., from a chat UI)
promptSummary	string?	Summary of the active prompt or context window

Pass agentContext as a parameter to startRun() and trackToolCall(). It is stored on runs and tool calls, and forwarded to approval requests for reviewer context.

import { AgentContext } from '@sequenceware/sdk';

TypeScript interfaces

All interfaces are exported from @sequenceware/sdk:

import type {
  ClientConfig,
  SequencewareEvent,
  AgentContext,
  ApprovalRequest,
  Policy,
  RunSummary,
  AuditLog,
} from '@sequenceware/sdk';

See the Client Reference for detailed interface definitions.