@namzu/sdk 0.1.4 → 0.1.5-rc.1

package/src/plugin/resolver.ts

@@ -1,14 +1,14 @@
 import { PLUGIN_NAMESPACE_SEPARATOR } from '../constants/plugin/index.js'
 import type { PluginRegistry } from '../registry/plugin/index.js'
-import type { ToolRegistry } from '../registry/tool/execute.js'
 import type { PluginId } from '../types/ids/index.js'
 import type { PluginContributionType } from '../types/plugin/index.js'
+import type { ToolRegistryContract } from '../types/tool/index.js'
 
 export class PluginResolver {
 	private pluginRegistry: PluginRegistry
-	private toolRegistry: ToolRegistry
+	private toolRegistry: ToolRegistryContract
 
-	constructor(pluginRegistry: PluginRegistry, toolRegistry: ToolRegistry) {
+	constructor(pluginRegistry: PluginRegistry, toolRegistry: ToolRegistryContract) {
 		this.pluginRegistry = pluginRegistry
 		this.toolRegistry = toolRegistry
 	}

package/src/registry/agent/definitions.ts

@@ -1,5 +1,6 @@
-import type { AgentDefinition, AgentType } from '../../types/agent/index.js'
-import type { Agent, BaseAgentConfig, BaseAgentResult } from '../../types/agent/index.js'
+import type { AgentType, BaseAgentConfig, BaseAgentResult } from '../../types/agent/base.js'
+import type { Agent } from '../../types/agent/core.js'
+import type { AgentDefinition } from '../../types/agent/factory.js'
 import { type Logger, getRootLogger } from '../../utils/logger.js'
 import { Registry } from '../Registry.js'
 

package/src/registry/tool/execute.ts

@@ -7,18 +7,15 @@ import type {
 	ToolAvailability,
 	ToolContext,
 	ToolDefinition,
+	ToolExecutionResult,
 	ToolRegistryConfig,
-	ToolResult,
 	ToolTierConfig,
 } from '../../types/tool/index.js'
 import { toErrorMessage } from '../../utils/error.js'
 import { type Logger, getRootLogger } from '../../utils/logger.js'
 import { Registry } from '../Registry.js'
 
-export interface ToolExecutionResult extends ToolResult {
-	permissionDenied?: boolean
-	permissionMessage?: string
-}
+export type { ToolExecutionResult }
 
 export class ToolRegistry extends Registry<ToolDefinition> {
 	private log: Logger

package/src/runtime/decision/parser.ts

@@ -17,14 +17,9 @@ export class DecisionParser {
 	parse(rawContent: string | null): DecisionParseResult {
 		if (!rawContent || rawContent.trim().length === 0) {
 			return {
-				ok: true,
-				decision: {
-					agentId: this.config.fallbackAgentId,
-					confidence: 0,
-					reasoning: 'Empty LLM response',
-				},
-				source: 'fallback',
-				reason: 'empty_response',
+				ok: false,
+				error: 'Empty LLM response',
+				rawContent: rawContent ?? '',
 			}
 		}
 
@@ -34,14 +29,9 @@ export class DecisionParser {
 				contentPreview: rawContent.slice(0, 200),
 			})
 			return {
-				ok: true,
-				decision: {
-					agentId: this.config.fallbackAgentId,
-					confidence: 0,
-					reasoning: 'Could not extract JSON from response',
-				},
-				source: 'fallback',
-				reason: 'json_extraction_failed',
+				ok: false,
+				error: 'Could not extract JSON from response',
+				rawContent,
 			}
 		}
 
@@ -54,14 +44,9 @@ export class DecisionParser {
 				contentPreview: jsonStr.slice(0, 200),
 			})
 			return {
-				ok: true,
-				decision: {
-					agentId: this.config.fallbackAgentId,
-					confidence: 0,
-					reasoning: `JSON parse failed: ${String(err)}`,
-				},
-				source: 'fallback',
-				reason: 'json_parse_failed',
+				ok: false,
+				error: `JSON parse failed: ${String(err)}`,
+				rawContent: jsonStr,
 			}
 		}
 
@@ -74,14 +59,9 @@ export class DecisionParser {
 			this.log.warn('Routing response failed schema validation', { errors })
 
 			return {
-				ok: true,
-				decision: {
-					agentId: this.config.fallbackAgentId,
-					confidence: 0,
-					reasoning: `Schema validation failed: ${errors}`,
-				},
-				source: 'fallback',
-				reason: 'schema_validation_failed',
+				ok: false,
+				error: `Schema validation failed: ${errors}`,
+				rawContent: jsonStr,
 			}
 		}
 
@@ -94,14 +74,9 @@ export class DecisionParser {
 			})
 
 			return {
-				ok: true,
-				decision: {
-					agentId: this.config.fallbackAgentId,
-					confidence: 0,
-					reasoning: `Unknown agentId "${response.agentId}", falling back`,
-				},
-				source: 'fallback',
-				reason: 'unknown_agent_id',
+				ok: false,
+				error: `Unknown agentId "${response.agentId}"`,
+				rawContent: jsonStr,
 			}
 		}
 

package/src/runtime/query/context-cache.ts

@@ -1,11 +1,10 @@
 import { createHash } from 'node:crypto'
-import type { ToolRegistry } from '../../registry/tool/execute.js'
 import type { AgentContextLevel } from '../../types/agent/factory.js'
 import type { ThreadId } from '../../types/ids/index.js'
 import type { AgentPersona } from '../../types/persona/index.js'
 import type { Skill } from '../../types/skills/index.js'
-import { PromptBuilder } from './prompt.js'
-import type { PromptSegments } from './prompt.js'
+import type { ToolRegistryContract } from '../../types/tool/index.js'
+import { PromptBuilder, type PromptSegments } from './prompt.js'
 
 export interface ContextCacheConfig {
 	agentId: string
@@ -17,7 +16,7 @@ export interface PromptCacheInput {
 	persona?: AgentPersona
 	skills?: Skill[]
 	basePrompt?: string
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	allowedTools?: string[]
 }
 

package/src/runtime/query/context.ts

@@ -10,8 +10,7 @@ import type { LLMProvider } from '../../types/provider/index.js'
 import type { AgentRunConfig } from '../../types/run/index.js'
 import type { ModelPricing } from '../../utils/cost.js'
 import { generateRunId } from '../../utils/id.js'
-import type { Logger } from '../../utils/logger.js'
-import { getRootLogger } from '../../utils/logger.js'
+import { type Logger, getRootLogger } from '../../utils/logger.js'
 
 export interface RunContextConfig {
 	agentId: string

package/src/runtime/query/events.ts

@@ -29,10 +29,10 @@ export class EventTranslator {
 	}
 
 	wireActivityStore(activityStore: ActivityStore, runId: RunId): void {
-		activityStore.on((event: ActivityEvent) => {
+		activityStore.on(async (event: ActivityEvent) => {
 			const activity = event.activity
 			if (event.type === 'activity.created') {
-				this.emitEvent({
+				await this.emitEvent({
 					type: 'activity_created',
 					runId,
 					activityId: activity.id,
@@ -40,7 +40,7 @@ export class EventTranslator {
 					description: activity.description,
 				})
 			} else {
-				this.emitEvent({
+				await this.emitEvent({
 					type: 'activity_updated',
 					runId,
 					activityId: activity.id,
@@ -53,13 +53,13 @@ export class EventTranslator {
 	}
 
 	wireTaskStore(taskStore: TaskStore, runId: RunId): () => void {
-		const unsubscribe = taskStore.on((event: TaskEvent) => {
+		const unsubscribe = taskStore.on(async (event: TaskEvent) => {
 			const task = event.task
 
 			if (task.runId !== runId) return
 			switch (event.type) {
 				case 'task.created':
-					this.emitEvent({
+					await this.emitEvent({
 						type: 'task_created',
 						runId,
 						taskId: task.id,
@@ -70,7 +70,7 @@ export class EventTranslator {
 				case 'task.updated':
 				case 'task.claimed':
 				case 'task.deleted':
-					this.emitEvent({
+					await this.emitEvent({
 						type: 'task_updated',
 						runId,
 						taskId: task.id,
@@ -89,11 +89,11 @@ export class EventTranslator {
 	}
 
 	wirePlanManager(planManager: PlanManager, runId: RunId): void {
-		planManager.on((event: PlanEvent) => {
+		planManager.on(async (event: PlanEvent) => {
 			const plan = event.plan
 			switch (event.type) {
 				case 'plan.ready':
-					this.emitEvent({
+					await this.emitEvent({
 						type: 'plan_ready',
 						runId,
 						planId: plan.id,
@@ -103,14 +103,14 @@ export class EventTranslator {
 					})
 					break
 				case 'plan.approved':
-					this.emitEvent({
+					await this.emitEvent({
 						type: 'plan_approved',
 						runId,
 						planId: plan.id,
 					})
 					break
 				case 'plan.rejected':
-					this.emitEvent({
+					await this.emitEvent({
 						type: 'plan_rejected',
 						runId,
 						planId: plan.id,
@@ -119,7 +119,7 @@ export class EventTranslator {
 					break
 				case 'plan.step_updated':
 					if (event.step) {
-						this.emitEvent({
+						await this.emitEvent({
 							type: 'plan_step_updated',
 							runId,
 							planId: plan.id,

package/src/runtime/query/executor.ts

@@ -1,27 +1,28 @@
 import { extractFromToolCall, extractFromToolResult } from '../../compaction/extractor.js'
 import type { WorkingStateManager } from '../../compaction/manager.js'
-import type { ToolRegistry } from '../../registry/tool/execute.js'
 import type { ActivityStore } from '../../store/activity/memory.js'
 import type { RunId } from '../../types/ids/index.js'
+import type { InvocationState } from '../../types/invocation/index.js'
 import { type Message, createToolMessage } from '../../types/message/index.js'
 import type { PermissionMode } from '../../types/permission/index.js'
 import type { ChatCompletionResponse } from '../../types/provider/index.js'
 import type { RunEvent } from '../../types/run/index.js'
 import type { Sandbox } from '../../types/sandbox/index.js'
-import type { ToolContext } from '../../types/tool/index.js'
+import type { ToolContext, ToolRegistryContract } from '../../types/tool/index.js'
 import type { Logger } from '../../utils/logger.js'
 import { compressShellOutput } from '../../utils/shell-compress.js'
 
 export type EmitEvent = (event: RunEvent) => Promise<void>
 
 export interface ToolExecutorConfig {
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	runId: RunId
 	workingDirectory: string
 	permissionMode: PermissionMode
 	env: Record<string, string>
 	abortSignal: AbortSignal
 	sandbox?: Sandbox
+	invocationState?: InvocationState
 }
 
 export interface ToolExecutionBatch {
@@ -91,6 +92,7 @@ export class ToolExecutor {
 				runId: this.config.runId,
 				workingDirectory: this.config.workingDirectory,
 			},
+			invocationState: this.config.invocationState,
 			toolRegistry: this.config.tools,
 			sandbox: this.config.sandbox,
 		}

package/src/runtime/query/index.ts

@@ -8,7 +8,6 @@ import { extractFromUserMessage } from '../../compaction/extractor.js'
 import { WorkingStateManager } from '../../compaction/manager.js'
 import type { CompactionConfig } from '../../config/runtime.js'
 import { getTracer } from '../../provider/telemetry/setup.js'
-import type { ToolRegistry } from '../../registry/tool/execute.js'
 import { GENAI, NAMZU, agentRunSpanName } from '../../telemetry/attributes.js'
 import { buildAdvisoryTools } from '../../tools/advisory/index.js'
 import { SearchToolsTool } from '../../tools/builtins/search-tools.js'
@@ -16,9 +15,13 @@ import { buildTaskTools } from '../../tools/task/index.js'
 import type { AdvisoryConfig } from '../../types/advisory/index.js'
 import type { RuntimeToolOverrides } from '../../types/agent/base.js'
 import type { AgentContextLevel } from '../../types/agent/factory.js'
-import type { CheckpointId, ResumeHandler } from '../../types/hitl/index.js'
-import { autoApproveHandler } from '../../types/hitl/index.js'
+import {
+	type CheckpointId,
+	type ResumeHandler,
+	autoApproveHandler,
+} from '../../types/hitl/index.js'
 import type { RunId, ThreadId } from '../../types/ids/index.js'
+import type { InvocationState } from '../../types/invocation/index.js'
 import { type Message, createSystemMessage } from '../../types/message/index.js'
 import type { AgentPersona } from '../../types/persona/index.js'
 import type { LLMProvider } from '../../types/provider/index.js'
@@ -27,6 +30,7 @@ import type { AgentRun, AgentRunConfig, RunEvent, RunEventListener } from '../..
 import type { Sandbox, SandboxProvider } from '../../types/sandbox/index.js'
 import type { Skill } from '../../types/skills/index.js'
 import type { TaskStore } from '../../types/task/index.js'
+import type { ToolRegistryContract } from '../../types/tool/index.js'
 import type { VerificationGateConfig } from '../../types/verification/index.js'
 import type { ModelPricing } from '../../utils/cost.js'
 import { VerificationGate } from '../../verification/gate.js'
@@ -47,7 +51,7 @@ export interface QueryParams {
 	skills?: Skill[]
 	basePrompt?: string
 	provider: LLMProvider
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	runConfig: AgentRunConfig
 	allowedTools?: string[]
 	agentId: string
@@ -102,6 +106,8 @@ export interface QueryParams {
 	pluginManager?: import('../../plugin/lifecycle.js').PluginLifecycleManager
 
 	sandboxProvider?: SandboxProvider
+
+	invocationState?: InvocationState
 }
 
 export async function* query(params: QueryParams): AsyncGenerator<RunEvent, AgentRun> {
@@ -186,6 +192,7 @@ export async function* query(params: QueryParams): AsyncGenerator<RunEvent, Agen
 			permissionMode: ctx.permissionMode,
 			env: params.runConfig.env ?? {},
 			abortSignal: ctx.abortController.signal,
+			invocationState: params.invocationState,
 		},
 		ctx.activityStore,
 		eventTranslator.emitEvent,

package/src/runtime/query/iteration/index.ts

@@ -6,13 +6,13 @@ import type { CompactionConfig } from '../../../config/runtime.js'
 import type { PlanManager } from '../../../manager/plan/lifecycle.js'
 import type { RunPersistence } from '../../../manager/run/persistence.js'
 import { getTracer } from '../../../provider/telemetry/setup.js'
-import type { ToolRegistry } from '../../../registry/tool/execute.js'
 import type { ActivityStore } from '../../../store/activity/memory.js'
 import { GENAI, NAMZU, agentIterationSpanName } from '../../../telemetry/attributes.js'
 import type { ResumeHandler } from '../../../types/hitl/index.js'
 import { createAssistantMessage, createUserMessage } from '../../../types/message/index.js'
 import type { LLMProvider } from '../../../types/provider/index.js'
 import type { AgentRunConfig, RunEvent, StopReason } from '../../../types/run/index.js'
+import type { ToolRegistryContract } from '../../../types/tool/index.js'
 import { toErrorMessage } from '../../../utils/error.js'
 import type { Logger } from '../../../utils/logger.js'
 import type { CheckpointManager } from '../checkpoint.js'
@@ -33,7 +33,7 @@ export type { ToolReviewOutcome } from './phases/index.js'
 export interface IterationConfig {
 	provider: LLMProvider
 	runConfig: AgentRunConfig
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	allowedTools?: string[]
 	taskGateway?: import('../../../types/agent/gateway.js').TaskGateway
 	taskStore?: import('../../../types/task/index.js').TaskStore

package/src/runtime/query/iteration/phases/advisory.ts

@@ -1,6 +1,5 @@
 import { serializeState } from '../../../../compaction/serializer.js'
-import type { AdvisoryRequest } from '../../../../types/advisory/index.js'
-import type { TriggerEvaluationState } from '../../../../types/advisory/index.js'
+import type { AdvisoryRequest, TriggerEvaluationState } from '../../../../types/advisory/index.js'
 import { createUserMessage } from '../../../../types/message/index.js'
 import type { ChatCompletionResponse } from '../../../../types/provider/index.js'
 import { toErrorMessage } from '../../../../utils/error.js'

package/src/runtime/query/iteration/phases/checkpoint.ts

@@ -1,7 +1,6 @@
 import type { RunEvent } from '../../../../types/run/index.js'
 import { CheckpointManager } from '../../checkpoint.js'
-import type { IterationContext, PhaseSignal } from './context.js'
-import { handleHITLDecision } from './context.js'
+import { type IterationContext, type PhaseSignal, handleHITLDecision } from './context.js'
 
 export async function* runIterationCheckpoint(
 	ctx: IterationContext,

package/src/runtime/query/iteration/phases/context.ts

@@ -4,7 +4,6 @@ import type { WorkingStateManager } from '../../../../compaction/manager.js'
 import type { CompactionConfig } from '../../../../config/runtime.js'
 import type { PlanManager } from '../../../../manager/plan/lifecycle.js'
 import type { RunPersistence } from '../../../../manager/run/persistence.js'
-import type { ToolRegistry } from '../../../../registry/tool/execute.js'
 import type { ActivityStore } from '../../../../store/activity/memory.js'
 import type { TaskGateway, TaskHandle } from '../../../../types/agent/gateway.js'
 import type { HITLResumeDecision, ResumeHandler } from '../../../../types/hitl/index.js'
@@ -12,6 +11,7 @@ import type { TaskId } from '../../../../types/ids/index.js'
 import type { LLMProvider } from '../../../../types/provider/index.js'
 import type { AgentRunConfig, RunEvent } from '../../../../types/run/index.js'
 import type { TaskStore } from '../../../../types/task/index.js'
+import type { ToolRegistryContract } from '../../../../types/tool/index.js'
 import type { Logger } from '../../../../utils/logger.js'
 import type { CheckpointManager } from '../../checkpoint.js'
 import type { EmitEvent } from '../../events.js'
@@ -27,7 +27,7 @@ export interface LaunchedTaskMeta {
 export interface IterationContext {
 	readonly provider: LLMProvider
 	readonly runConfig: AgentRunConfig
-	readonly tools: ToolRegistry
+	readonly tools: ToolRegistryContract
 	readonly allowedTools?: string[]
 	readonly runMgr: RunPersistence
 	readonly toolExecutor: ToolExecutor

package/src/runtime/query/iteration/phases/plan.ts

@@ -1,6 +1,5 @@
 import type { RunEvent } from '../../../../types/run/index.js'
-import type { IterationContext, PhaseSignal } from './context.js'
-import { handleHITLDecision } from './context.js'
+import { type IterationContext, type PhaseSignal, handleHITLDecision } from './context.js'
 
 export async function* runPlanGate(ctx: IterationContext): AsyncGenerator<RunEvent, PhaseSignal> {
 	if (!ctx.planManager.active || ctx.planManager.active.status !== 'ready') {

package/src/runtime/query/prompt.ts

@@ -1,9 +1,9 @@
 import { FILESYSTEM_TOOLS } from '../../constants/tools/index.js'
 import { assembleSystemPrompt } from '../../persona/assembler.js'
-import type { ToolRegistry } from '../../registry/tool/execute.js'
 import type { AgentContextLevel } from '../../types/agent/factory.js'
 import type { AgentPersona } from '../../types/persona/index.js'
 import type { Skill } from '../../types/skills/index.js'
+import type { ToolRegistryContract } from '../../types/tool/index.js'
 
 export interface PromptSegments {
 	/** Layers 1-6: basePrompt, persona identity/expertise/reflexes/skills/outputDiscipline. Stable within a run. */
@@ -21,7 +21,7 @@ export interface PromptBuilderConfig {
 
 	basePrompt?: string
 
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	allowedTools?: string[]
 }
 
@@ -34,7 +34,7 @@ Platform: ${process.platform}
 IMPORTANT: Always use absolute paths based on the working directory above. Before reading a file, use the glob tool to discover actual file paths — never guess or hallucinate paths.`
 }
 
-function hasFilesystemTools(tools: ToolRegistry, allowedTools?: string[]): boolean {
+function hasFilesystemTools(tools: ToolRegistryContract, allowedTools?: string[]): boolean {
 	const activeTools = allowedTools ?? tools.listNames()
 	return activeTools.some((name) => FILESYSTEM_TOOLS.has(name))
 }

package/src/runtime/query/result.ts

@@ -1,5 +1,4 @@
-import { SpanStatusCode } from '@opentelemetry/api'
-import type { Span } from '@opentelemetry/api'
+import { type Span, SpanStatusCode } from '@opentelemetry/api'
 import type { PlanManager } from '../../manager/plan/lifecycle.js'
 import type { RunPersistence } from '../../manager/run/persistence.js'
 import type { ActivityStore } from '../../store/activity/memory.js'

package/src/runtime/query/tooling.ts

@@ -1,20 +1,22 @@
-import type { ToolRegistry } from '../../registry/tool/execute.js'
 import type { ActivityStore } from '../../store/activity/memory.js'
 import type { RunId } from '../../types/ids/index.js'
+import type { InvocationState } from '../../types/invocation/index.js'
 import type { PermissionMode } from '../../types/permission/index.js'
 import type { RunEvent } from '../../types/run/index.js'
+import type { ToolRegistryContract } from '../../types/tool/index.js'
 import type { Logger } from '../../utils/logger.js'
 import { ToolExecutor } from './executor.js'
 
 export type EmitEvent = (event: RunEvent) => Promise<void>
 
 export interface ToolingBootstrapConfig {
-	tools: ToolRegistry
+	tools: ToolRegistryContract
 	runId: RunId
 	workingDirectory: string
 	permissionMode: PermissionMode
 	env: Record<string, string>
 	abortSignal: AbortSignal
+	invocationState?: InvocationState
 }
 
 export class ToolingBootstrap {
@@ -32,6 +34,7 @@ export class ToolingBootstrap {
 				permissionMode: config.permissionMode,
 				env: config.env,
 				abortSignal: config.abortSignal,
+				invocationState: config.invocationState,
 			},
 			activityStore,
 			emitEvent,

package/src/store/conversation/memory.ts

@@ -1,7 +1,11 @@
+import { findSafeTrimIndex } from '../../compaction/dangling.js'
 import type { ConversationStore } from '../../types/conversation/index.js'
 import type { MessageId, RunId, ThreadId } from '../../types/ids/index.js'
-import type { Message } from '../../types/message/index.js'
-import { createAssistantMessage, createUserMessage } from '../../types/message/index.js'
+import {
+	type Message,
+	createAssistantMessage,
+	createUserMessage,
+} from '../../types/message/index.js'
 import { extractFinalResponse } from '../../utils/conversation.js'
 import { generateMessageId } from '../../utils/id.js'
 
@@ -35,16 +39,28 @@ export class InMemoryConversationStore implements ConversationStore {
 		const messages = this.threads.get(threadId)
 		if (!messages) return []
 
-		const slice = messages.length > this.maxMessages ? messages.slice(-this.maxMessages) : messages
-
-		return slice.map((m) => {
+		// Convert to Message type for trim safety check
+		const convertedMessages: Message[] = messages.map((m) => {
 			switch (m.role) {
 				case 'user':
 					return createUserMessage(m.content)
 				case 'assistant':
 					return createAssistantMessage(m.content)
+				default: {
+					const _exhaustive: never = m.role
+					throw new Error(`Unhandled message role: ${_exhaustive}`)
+				}
 			}
 		})
+
+		// Apply safe trimming to preserve tool call/result atomicity
+		let trimStartIdx = 0
+		if (convertedMessages.length > this.maxMessages) {
+			const desiredTrimPoint = convertedMessages.length - this.maxMessages
+			trimStartIdx = findSafeTrimIndex(convertedMessages, desiredTrimPoint)
+		}
+
+		return convertedMessages.slice(trimStartIdx)
 	}
 
 	persistRunResult(threadId: ThreadId, runId: RunId, messages: Message[]): void {

package/src/store/run/disk.ts

@@ -1,8 +1,7 @@
 import { appendFile, mkdir, readFile, readdir, rename, unlink, writeFile } from 'node:fs/promises'
 import { join } from 'node:path'
 import type { CheckpointId, IterationCheckpoint } from '../../types/hitl/index.js'
-import type { AgentRun, RunEvent } from '../../types/run/index.js'
-import type { RunStoreConfig } from '../../types/run/index.js'
+import type { AgentRun, RunEvent, RunStoreConfig } from '../../types/run/index.js'
 import { type Logger, getRootLogger } from '../../utils/logger.js'
 
 export class RunDiskStore {
@@ -96,8 +95,9 @@ export class RunDiskStore {
 		try {
 			const content = await readFile(join(dir, 'checkpoints', `${checkpointId}.json`), 'utf-8')
 			return JSON.parse(content) as IterationCheckpoint
-		} catch {
-			return null
+		} catch (err) {
+			if (isFileNotFound(err)) return null
+			throw err
 		}
 	}
 
@@ -117,8 +117,9 @@ export class RunDiskStore {
 				}
 			}
 			return checkpoints.sort((a, b) => a.createdAt - b.createdAt)
-		} catch {
-			return []
+		} catch (err) {
+			if (isFileNotFound(err)) return []
+			throw err
 		}
 	}
 
@@ -126,7 +127,9 @@ export class RunDiskStore {
 		const dir = this.requireInit()
 		try {
 			await unlink(join(dir, 'checkpoints', `${checkpointId}.json`))
-		} catch {}
+		} catch (err) {
+			if (!isFileNotFound(err)) throw err
+		}
 	}
 
 	static async listRuns(baseDir: string): Promise<
@@ -142,8 +145,9 @@ export class RunDiskStore {
 			const indexPath = join(baseDir, 'index.json')
 			const content = await readFile(indexPath, 'utf-8')
 			return JSON.parse(content)
-		} catch {
-			return []
+		} catch (err) {
+			if (isFileNotFound(err)) return []
+			throw err
 		}
 	}
 
@@ -166,13 +170,7 @@ export class RunDiskStore {
 				const content = await readFile(indexPath, 'utf-8')
 				index = JSON.parse(content)
 			} catch (err) {
-				const isNotFound =
-					typeof err === 'object' &&
-					err !== null &&
-					(err as NodeJS.ErrnoException).code === 'ENOENT'
-				if (!isNotFound) {
-					this.log.warn('Failed to read run index — starting fresh', { error: String(err) })
-				}
+				if (!isFileNotFound(err)) throw err
 			}
 
 			const entry = {
@@ -216,6 +214,10 @@ async function atomicWriteJson(filePath: string, value: unknown): Promise<void>
 	await atomicWriteFile(filePath, JSON.stringify(value, null, 2))
 }
 
+function isFileNotFound(err: unknown): boolean {
+	return typeof err === 'object' && err !== null && (err as NodeJS.ErrnoException).code === 'ENOENT'
+}
+
 export const SessionStore = RunDiskStore
 
 export type SessionStore = RunDiskStore