@poncho-ai/harness 0.60.1 → 0.61.1

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.60.1 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
+> @poncho-ai/harness@0.61.1 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
 > node scripts/embed-docs.js && tsup src/index.ts --format esm --dts
 
 [embed-docs] Generated poncho-docs.ts with 4 topics
@@ -8,9 +8,9 @@
 CLI tsup v8.5.1
 CLI Target: es2022
 ESM Build start
-ESM dist/index.js            574.16 KB
+ESM dist/index.js            580.50 KB
 ESM dist/isolate-F2PPSUL6.js 53.82 KB
-ESM ⚡️ Build success in 245ms
+ESM ⚡️ Build success in 227ms
 DTS Build start
-DTS ⚡️ Build success in 7563ms
-DTS dist/index.d.ts 105.62 KB
+DTS ⚡️ Build success in 7778ms
+DTS dist/index.d.ts 110.56 KB

package/CHANGELOG.md

@@ -1,5 +1,53 @@
 # @poncho-ai/harness
 
+## 0.61.1
+
+### Patch Changes
+
+- [`a16f4fa`](https://github.com/cesr/poncho-ai/commit/a16f4fa32f0809d6ff73f0041cb0e0a3fb980bb1) Thanks [@cesr](https://github.com/cesr)! - Fix: checkpoint-resume no longer drops a turn's user message from the model
+  transcript. Extract the checkpoint→transcript assembly the orchestrator's
+  `resumeRunFromCheckpoint` used into shared, exported helpers —
+  `assembleCheckpointMessages`, `buildToolResultMessage`, `buildResumeCheckpoints`
+  — so external embedders (which execute gated tools themselves) reconstruct the
+  canonical transcript identically instead of re-deriving the index arithmetic
+  and drifting. Add a transcript-integrity guard in `applyTurnMetadata` that
+  logs when a finalize would leave the latest user message out of the
+  model-facing transcript.
+
+## 0.61.0
+
+### Minor Changes
+
+- [#192](https://github.com/cesr/poncho-ai/pull/192) [`bfd8012`](https://github.com/cesr/poncho-ai/commit/bfd80125ad71cd4401221c598faebf71afda2078) Thanks [@cesr](https://github.com/cesr)! - Fix context-compaction correctness so the agent stops losing context and
+  mislabeling failed work after a compaction:
+  - **Turn-based retention.** Compaction now preserves the last N whole _turns_
+    verbatim (new `compaction.keepRecentTurns`, default 4) instead of N messages,
+    which in tool-heavy turns collapsed to just the summary. The preserved side is
+    bounded by a token budget (≤ 50% of the context window) so keeping recent turns
+    can't leave the post-compaction context above the trigger (re-compaction
+    thrash / overflow). Adds exported `findSafeSplitPointByTurns`.
+  - **Faithful summaries.** The summarization output cap is raised 768 → 8192
+    tokens (768 physically truncated summaries mid-content); per-message truncation
+    1200 → 4000 chars, with a total summarizer-input budget that drops the oldest
+    non-error messages first. The prompt now requires a non-omittable "Unresolved
+    errors & failures" section, a "Pending promises" section, forbids claiming
+    unconfirmed completion, and preserves identifiers verbatim.
+  - **Structured subagent task outcome.** New `PendingSubagentResult.taskOutcome`
+    (`succeeded | failed | partial | unknown`) distinct from run status: a subagent
+    that ran but failed its task is no longer recorded as "completed". Subagents
+    self-report a machine-readable verdict; it is parsed deterministically
+    (defaulting to "unknown", never success) and rendered in the callback header
+    and compaction ledger. The subagent digest is enlarged (2000 chars, ungated on
+    status) so the failure reason survives compaction.
+  - **Per-run `maxSteps` override.** New `RunInput.maxSteps` lets a caller raise the
+    step ceiling for foreground turns without raising it for background/job turns
+    that share the same agent definition.
+
+### Patch Changes
+
+- Updated dependencies [[`bfd8012`](https://github.com/cesr/poncho-ai/commit/bfd80125ad71cd4401221c598faebf71afda2078)]:
+  - @poncho-ai/sdk@1.17.0
+
 ## 0.60.1
 
 ### Patch Changes

package/dist/index.d.ts

@@ -43,6 +43,16 @@ interface AgentFrontmatter {
 interface CompactionConfig {
     enabled: boolean;
     trigger: number;
+    /**
+     * Primary retention unit: number of whole recent *turns* (a user message plus
+     * the assistant/tool messages that follow it) preserved verbatim after the
+     * summary. See `findSafeSplitPointByTurns`.
+     */
+    keepRecentTurns: number;
+    /**
+     * Fallback retention floor, in messages, used only for the single-giant-turn
+     * case where no turn boundary yields a safe split.
+     */
     keepRecentMessages: number;
     instructions?: string;
 }
@@ -105,8 +115,37 @@ declare const estimateTotalTokens: (systemPrompt: string, messages: Message[], t
  * Returns -1 if no valid split point is found.
  */
 declare const findSafeSplitPoint: (messages: Message[], keepRecentMessages: number) => number;
+/**
+ * Find the split index that preserves the last `keepRecentTurns` whole *turns*
+ * verbatim. A turn begins at a `role:"user"` message and runs until the next
+ * `user` message. Everything before the split is folded into the summary;
+ * everything from the split onward is kept as-is.
+ *
+ * Two guards make a candidate split acceptable:
+ *  1. The compacted side has at least `MIN_COMPACTABLE_MESSAGES` messages.
+ *  2. The preserved side estimates at most `maxPreservedTokens` — so keeping N
+ *     large turns can never leave the post-compaction context above the
+ *     compaction trigger (which caused re-compaction thrash / window overflow).
+ * The tool-call-orphan guard (`splitOrphansToolCalls`) is preserved: a split
+ * that would strand an assistant tool-call on the compacted side steps to an
+ * earlier user boundary (which only preserves more, always safe).
+ *
+ * We try the largest N (up to `keepRecentTurns`) first and decrement, so we
+ * keep as many turns as fit. If no turn boundary yields a safe split (e.g. a
+ * single giant turn near the window), fall back to the message-based
+ * `findSafeSplitPoint` so the middle of that turn still compacts.
+ *
+ * Returns -1 when nothing safe can be compacted.
+ */
+declare const findSafeSplitPointByTurns: (messages: Message[], keepRecentTurns: number, keepRecentMessagesFallback: number, maxPreservedTokens: number) => number;
 interface CompactMessagesOptions {
     instructions?: string;
+    /**
+     * The model's context window (tokens). Used to bound the preserved side of a
+     * turn-based split so keeping N recent turns can't leave the post-compaction
+     * context above the compaction trigger. Defaults to 200k when omitted.
+     */
+    contextWindow?: number;
 }
 interface CompactResult {
     compacted: boolean;
@@ -207,10 +246,22 @@ interface StateStore {
     set(state: ConversationState): Promise<void>;
     delete(runId: string): Promise<void>;
 }
+/**
+ * Whether a subagent actually accomplished its task — distinct from the *run*
+ * status (`status` below), which is "completed" whenever the run merely
+ * finished. A subagent that ran to completion but could not do the task (e.g.
+ * lacked the required tools) is `runStatus:"completed"` but `taskOutcome:
+ * "failed"`. Derived at subagent finalize; "unknown" when it can't be
+ * determined. Consumed by the compaction subagent ledger so a failed task can
+ * never be summarized as "completed".
+ */
+type SubagentTaskOutcome = "succeeded" | "failed" | "partial" | "unknown";
 interface PendingSubagentResult {
     subagentId: string;
     task: string;
     status: "completed" | "error" | "stopped";
+    /** Did the subagent accomplish its task? See {@link SubagentTaskOutcome}. */
+    taskOutcome: SubagentTaskOutcome;
     result?: _poncho_ai_sdk.RunResult;
     error?: _poncho_ai_sdk.AgentFailure;
     timestamp: number;
@@ -2171,6 +2222,40 @@ declare const buildApprovalCheckpoints: ({ approvals, runId, checkpointMessages,
     pendingToolCalls: PendingToolCall[];
     kind?: "approval" | "device";
 }) => NonNullable<Conversation["pendingApprovals"]>;
+/** Reconstruct the full canonical transcript a checkpoint resumes from: the
+ *  prior history (before the checkpointed turn) + the checkpoint delta.
+ *  Handles both storage conventions via `normalizeApprovalCheckpoint` — the
+ *  initial checkpoint (base = prior length, delta) and a resume-created one
+ *  (base 0, full canonical). */
+declare const assembleCheckpointMessages: (conversation: Conversation, checkpoint: StoredApproval) => Message[];
+/** Pair an assistant tool-call message's `tool_calls` with externally-computed
+ *  results into the single `role:"tool"` message a continuation reads.
+ *  Returns undefined when `assistantMsg` isn't an assistant message with
+ *  parseable tool calls. Used to persist the resume's canonical history so it
+ *  matches what `continueFromToolResult` fed the model. Missing results
+ *  default to the deferred-error marker (same text the run uses). */
+declare const buildToolResultMessage: (assistantMsg: Message | undefined, toolResults: Array<{
+    callId: string;
+    toolName: string;
+    result?: unknown;
+    error?: string;
+}>) => Message | undefined;
+/** Build the `pendingApprovals` rows for a checkpoint reached DURING a resume
+ *  continuation. Stores the WHOLE canonical history the continuation ran with
+ *  (`priorMessages` = prior + tool result + new delta) with
+ *  `baseMessageCount: 0`. Keeping the full history here — rather than a delta
+ *  + a base count into a different message array — is what lets the next
+ *  resume reconstruct with no index arithmetic. */
+declare const buildResumeCheckpoints: ({ priorMessages, checkpointEvent, runId, kind, }: {
+    priorMessages: Message[];
+    checkpointEvent: {
+        approvals: ApprovalEventItem[];
+        checkpointMessages: Message[];
+        pendingToolCalls: PendingToolCall[];
+    };
+    runId: string;
+    kind?: "approval" | "device";
+}) => NonNullable<Conversation["pendingApprovals"]>;
 declare const applyTurnMetadata: (conv: Conversation, meta: TurnResultMetadata, opts?: {
     clearContinuation?: boolean;
     clearApprovals?: boolean;
@@ -2404,6 +2489,12 @@ interface RunConversationTurnOpts {
      * built with an OTLP exporter attached.
      */
     suppressTelemetry?: boolean;
+    /**
+     * Forwarded to `RunInput.maxSteps`: per-run step-ceiling override. Lets the
+     * foreground chat turn run a higher ceiling than background/job turns that
+     * share the same agent definition.
+     */
+    maxSteps?: number;
     /**
      * Forwarded to `RunInput.telemetryAttributes` — extra attributes for the
      * `invoke_agent` root span (e.g. run kind / job name for segmentation).
@@ -2446,4 +2537,4 @@ type NewEntryNoId = NewConversationEntry extends infer T ? T extends NewConversa
  */
 declare const appendEntriesSafe: (store: ConversationStore, conversation: Pick<Conversation, "conversationId" | "ownerId" | "tenantId">, entries: NewEntryNoId[], log: Logger) => Promise<ConversationEntry[]>;
 
-export { type ActiveConversationRun, type ActiveSubagentRun, type AgentFrontmatter, AgentHarness, type AgentIdentity, type AgentLimitsConfig, type AgentModelConfig, AgentOrchestrator, type ApprovalEventItem, type ArchivedToolResult$1 as ArchivedToolResult, type BashConfig, BashEnvironmentManager, type BashExecutionLimits, type BuiltInToolToggles, CALLBACK_LOCK_STALE_MS, type CallbackStartedEntry, type CompactMessagesOptions, type CompactResult, type CompactionConfig, type ContinuationHooks, type Conversation, type ConversationCreateInit, type ConversationEntry, type ConversationState, type ConversationStatusSnapshot, type ConversationStore, type ConversationSummary, type CreateSkillToolsOptions, type CronJobConfig, DEFAULT_AGENT_DESCRIPTION, DEFAULT_AGENT_NAME, DEFAULT_MAX_STEPS, DEFAULT_MODEL_NAME, DEFAULT_MODEL_PROVIDER, DEFAULT_TAGLINE, DEFAULT_TEMPERATURE, DEFAULT_TIMEOUT, type DefaultAgentDefinitionOptions, type EventSink, type ExecuteTurnResult, type HarnessOptions, type HarnessRunOutput, type HistorySource, InMemoryConversationStore, InMemoryEngine, InMemoryStateStore, type IsolateBinding, type IsolateConfig, LocalMcpBridge, LocalUploadStore, MAX_CONCURRENT_SUBAGENTS, MAX_CONTINUATION_COUNT, MAX_SUBAGENT_CALLBACK_COUNT, MAX_SUBAGENT_NESTING, type MainMemory, type McpConfig, type MemoryConfig, type MemoryStore, type MessagingChannelConfig, type ModelProviderFactory, type MountProvider, type NetworkConfig, type NewConversationEntry, OPENAI_CODEX_CLIENT_ID, type OpenAICodexAuthConfig, type OpenAICodexDeviceAuthRequest, type OpenAICodexSession, type OrchestratorHooks, type OrchestratorOptions, type OtlpConfig, type OtlpOption, PONCHO_UPLOAD_SCHEME, type ParsedAgent, type PendingSubagentApproval, type PendingSubagentResult, type PendingToolCall, type PonchoConfig, PonchoFsAdapter, PostgresEngine, type ProviderConfig, type Recurrence, type RecurrenceType, type Reminder, type ReminderCreateInput, type ReminderStatus, type ReminderStore, type RemoteMcpServerConfig, type RunConversationTurnOpts, type RunConversationTurnResult, type RunOutcome, type RunRequest, type RuntimeRenderContext, S3UploadStore, STALE_SUBAGENT_THRESHOLD_MS, STORAGE_SCHEMA_VERSION, type SecretsStore, type SkillContextEntry, type SkillMetadata, type SkillSource, SqliteEngine, type StateConfig, type StateProviderName, type StateStore, type StorageConfig, type StorageEngine, type StorageFactoryOptions, type StorageProvider, type StoredApproval, type SubagentManager, type SubagentResult, type SubagentResultEntry, type SubagentSpawnResult, type SubagentSummary, type SubagentTranscript, type SubagentTranscriptMode, TOOL_RESULT_ARCHIVE_PARAM, type TelemetryConfig, TelemetryEmitter, type TenantTokenPayload, type ToolAccess, type ToolCall, ToolDispatcher, type ToolExecutionResult, type TurnDraftState, type TurnResultMetadata, type TurnSection, type UploadStore, type UploadsConfig, VFS_SCHEME, VercelBlobUploadStore, type VfsDirEntry, type VfsStat, type VirtualMount, abnormalEndResponse, appendEntriesSafe, applyTurnMetadata, buildAgentDirectoryName, buildApprovalCheckpoints, buildAssistantMetadata, buildSkillContextWindow, buildToolCompletedText, cloneSections, compactMessages, completeOpenAICodexDeviceAuth, computeNextOccurrence, createBashTool, createConversationStore, createConversationStoreFromEngine, createDefaultTools, createDeleteDirectoryTool, createDeleteTool, createEditTool, createMemoryStore, createMemoryStoreFromEngine, createMemoryTools, createModelProvider, createReminderStore, createReminderStoreFromEngine, createReminderTools, createSearchTools, createSecretsStore, createSkillTools, createStateStore, createStorageEngine, createSubagentTools, createTodoStoreFromEngine, createTurnDraftState, createUploadStore, createWriteTool, decodeFileInputData, defaultAgentDefinition, deleteOpenAICodexSession, deriveUploadKey, ensureAgentIdentity, estimateTokens, estimateTotalTokens, executeConversationTurn, findSafeSplitPoint, flushTurnDraft, generateAgentId, getAgentStoreDirectory, getModelContextWindow, getOpenAICodexAccessToken, getOpenAICodexAuthFilePath, getOpenAICodexRequiredScopes, getPendingSubagentResults, getPonchoStoreRoot, isMessageArray, jsonSchemaToZod, lastAssistantText, loadCanonicalHistory, loadPonchoConfig, loadRunHistory, loadSkillContext, loadSkillInstructions, loadSkillMetadata, loadSkillMetadataFromDirs, loadVfsSkillMetadata, mergeSkills, normalizeApprovalCheckpoint, normalizeOtlp, normalizeScriptPolicyPath, normalizeToolAccess, parseAgentFile, parseAgentMarkdown, parseSkillFrontmatter, ponchoDocsTool, readOpenAICodexSession, readSkillResource, realResponseText, recordStandardTurnEvent, renderAgentPrompt, resolveAgentIdentity, resolveCompactionConfig, resolveEnv, resolveMemoryConfig, resolveRunRequest, resolveSkillDirs, resolveStateConfig, runConversationTurn, slugifyStorageComponent, startOpenAICodexDeviceAuth, stripPillMetaTokens, verifyTenantToken, withToolResultArchiveParam, writeOpenAICodexSession };
+export { type ActiveConversationRun, type ActiveSubagentRun, type AgentFrontmatter, AgentHarness, type AgentIdentity, type AgentLimitsConfig, type AgentModelConfig, AgentOrchestrator, type ApprovalEventItem, type ArchivedToolResult$1 as ArchivedToolResult, type BashConfig, BashEnvironmentManager, type BashExecutionLimits, type BuiltInToolToggles, CALLBACK_LOCK_STALE_MS, type CallbackStartedEntry, type CompactMessagesOptions, type CompactResult, type CompactionConfig, type ContinuationHooks, type Conversation, type ConversationCreateInit, type ConversationEntry, type ConversationState, type ConversationStatusSnapshot, type ConversationStore, type ConversationSummary, type CreateSkillToolsOptions, type CronJobConfig, DEFAULT_AGENT_DESCRIPTION, DEFAULT_AGENT_NAME, DEFAULT_MAX_STEPS, DEFAULT_MODEL_NAME, DEFAULT_MODEL_PROVIDER, DEFAULT_TAGLINE, DEFAULT_TEMPERATURE, DEFAULT_TIMEOUT, type DefaultAgentDefinitionOptions, type EventSink, type ExecuteTurnResult, type HarnessOptions, type HarnessRunOutput, type HistorySource, InMemoryConversationStore, InMemoryEngine, InMemoryStateStore, type IsolateBinding, type IsolateConfig, LocalMcpBridge, LocalUploadStore, MAX_CONCURRENT_SUBAGENTS, MAX_CONTINUATION_COUNT, MAX_SUBAGENT_CALLBACK_COUNT, MAX_SUBAGENT_NESTING, type MainMemory, type McpConfig, type MemoryConfig, type MemoryStore, type MessagingChannelConfig, type ModelProviderFactory, type MountProvider, type NetworkConfig, type NewConversationEntry, OPENAI_CODEX_CLIENT_ID, type OpenAICodexAuthConfig, type OpenAICodexDeviceAuthRequest, type OpenAICodexSession, type OrchestratorHooks, type OrchestratorOptions, type OtlpConfig, type OtlpOption, PONCHO_UPLOAD_SCHEME, type ParsedAgent, type PendingSubagentApproval, type PendingSubagentResult, type PendingToolCall, type PonchoConfig, PonchoFsAdapter, PostgresEngine, type ProviderConfig, type Recurrence, type RecurrenceType, type Reminder, type ReminderCreateInput, type ReminderStatus, type ReminderStore, type RemoteMcpServerConfig, type RunConversationTurnOpts, type RunConversationTurnResult, type RunOutcome, type RunRequest, type RuntimeRenderContext, S3UploadStore, STALE_SUBAGENT_THRESHOLD_MS, STORAGE_SCHEMA_VERSION, type SecretsStore, type SkillContextEntry, type SkillMetadata, type SkillSource, SqliteEngine, type StateConfig, type StateProviderName, type StateStore, type StorageConfig, type StorageEngine, type StorageFactoryOptions, type StorageProvider, type StoredApproval, type SubagentManager, type SubagentResult, type SubagentResultEntry, type SubagentSpawnResult, type SubagentSummary, type SubagentTaskOutcome, type SubagentTranscript, type SubagentTranscriptMode, TOOL_RESULT_ARCHIVE_PARAM, type TelemetryConfig, TelemetryEmitter, type TenantTokenPayload, type ToolAccess, type ToolCall, ToolDispatcher, type ToolExecutionResult, type TurnDraftState, type TurnResultMetadata, type TurnSection, type UploadStore, type UploadsConfig, VFS_SCHEME, VercelBlobUploadStore, type VfsDirEntry, type VfsStat, type VirtualMount, abnormalEndResponse, appendEntriesSafe, applyTurnMetadata, assembleCheckpointMessages, buildAgentDirectoryName, buildApprovalCheckpoints, buildAssistantMetadata, buildResumeCheckpoints, buildSkillContextWindow, buildToolCompletedText, buildToolResultMessage, cloneSections, compactMessages, completeOpenAICodexDeviceAuth, computeNextOccurrence, createBashTool, createConversationStore, createConversationStoreFromEngine, createDefaultTools, createDeleteDirectoryTool, createDeleteTool, createEditTool, createMemoryStore, createMemoryStoreFromEngine, createMemoryTools, createModelProvider, createReminderStore, createReminderStoreFromEngine, createReminderTools, createSearchTools, createSecretsStore, createSkillTools, createStateStore, createStorageEngine, createSubagentTools, createTodoStoreFromEngine, createTurnDraftState, createUploadStore, createWriteTool, decodeFileInputData, defaultAgentDefinition, deleteOpenAICodexSession, deriveUploadKey, ensureAgentIdentity, estimateTokens, estimateTotalTokens, executeConversationTurn, findSafeSplitPoint, findSafeSplitPointByTurns, flushTurnDraft, generateAgentId, getAgentStoreDirectory, getModelContextWindow, getOpenAICodexAccessToken, getOpenAICodexAuthFilePath, getOpenAICodexRequiredScopes, getPendingSubagentResults, getPonchoStoreRoot, isMessageArray, jsonSchemaToZod, lastAssistantText, loadCanonicalHistory, loadPonchoConfig, loadRunHistory, loadSkillContext, loadSkillInstructions, loadSkillMetadata, loadSkillMetadataFromDirs, loadVfsSkillMetadata, mergeSkills, normalizeApprovalCheckpoint, normalizeOtlp, normalizeScriptPolicyPath, normalizeToolAccess, parseAgentFile, parseAgentMarkdown, parseSkillFrontmatter, ponchoDocsTool, readOpenAICodexSession, readSkillResource, realResponseText, recordStandardTurnEvent, renderAgentPrompt, resolveAgentIdentity, resolveCompactionConfig, resolveEnv, resolveMemoryConfig, resolveRunRequest, resolveSkillDirs, resolveStateConfig, runConversationTurn, slugifyStorageComponent, startOpenAICodexDeviceAuth, stripPillMetaTokens, verifyTenantToken, withToolResultArchiveParam, writeOpenAICodexSession };