@poncho-ai/harness 0.39.0 → 0.40.0

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.39.0 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
+> @poncho-ai/harness@0.40.0 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            479.07 KB
-ESM dist/isolate-TCWTUVG4.js 47.34 KB
-ESM ⚡️ Build success in 251ms
+ESM dist/index.js            489.99 KB
+ESM dist/isolate-VY35DGLM.js 49.43 KB
+ESM ⚡️ Build success in 173ms
 DTS Build start
-DTS ⚡️ Build success in 8269ms
-DTS dist/index.d.ts 73.95 KB
+DTS ⚡️ Build success in 7364ms
+DTS dist/index.d.ts 75.12 KB

package/CHANGELOG.md

@@ -1,5 +1,150 @@
 # @poncho-ai/harness
 
+## 0.40.0
+
+### Minor Changes
+
+- [#105](https://github.com/cesr/poncho-ai/pull/105) [`e127174`](https://github.com/cesr/poncho-ai/commit/e12717415b1114c5e9a58e7c51fcf9e038218f9f) Thanks [@cesr](https://github.com/cesr)! - feat: tenant-authored skills in the VFS
+
+  Tenants can now author skills in their VFS at `/skills/<name>/SKILL.md`
+  (plus sibling files such as `scripts/*.ts` and `references/*.md`). VFS
+  skills are merged with the agent's repo skills per-tenant when building
+  the `<available_skills>` block in the system prompt; repo skills win on
+  name collision (a warning is logged for the dropped VFS skill).
+
+  VFS skills can ship runnable scripts in their tree (`scripts/foo.ts`
+  etc.); the agent runs them via the existing `run_code` tool with
+  `file: "/skills/<name>/scripts/foo.ts"`, which executes in the sandboxed
+  isolated-vm runtime. `run_skill_script` remains for repo-shipped skills
+  only (jiti, full Node access), and returns a clear redirect when
+  called against a VFS skill. The agent's tool-policy lookups still
+  resolve against repo skills only, so tenants cannot grant themselves
+  new MCP tools by uploading a SKILL.md (security boundary).
+
+  `run_code` is enhanced so skill-authored scripts feel natural:
+  - Accepts top-level `export const run = ...`, `export default function ...`,
+    and `export default <expr>;` (the keyword is stripped at strip-TypeScript
+    time; `export default <expr>` becomes a `__default` binding).
+  - New optional `input` parameter, exposed inside the script as the global
+    `__input`.
+  - If the script defines a top-level `run` / `default` / `main` / `handler`
+    function and doesn't `return` on its own, the dispatcher invokes that
+    function with `__input` and returns its result. Existing
+    return-style scripts are unaffected.
+
+  The CLI Files sidebar already exposes the VFS, so creating a tenant
+  skill is just writing to `/skills/...` from the UI or via the agent's
+  own VFS write tools — the harness invalidates its per-tenant skill
+  cache on writes under `/skills/`.
+
+### Patch Changes
+
+- [`d24c152`](https://github.com/cesr/poncho-ai/commit/d24c152c1ecb9bfe59b086cb1f18a5ab43688223) Thanks [@cesr](https://github.com/cesr)! - fix(harness): cap `_toolResultArchive` size per conversation, FIFO-evict oldest
+
+  Heap-snapshot evidence from a 3.7 GB OOM showed 147,448 retained strings,
+  including 8 exact duplicates (~239 KB each) of the same browser-extracted
+  page text. The browser screenshot/snapshot skip-list from a prior fix
+  didn't help because page-text/web-extract tools still archived their
+  full payloads in `_toolResultArchive`, with no eviction across the
+  session.
+
+  Add a per-conversation archive byte cap (default 25 MB, configurable via
+  `PONCHO_TOOL_ARCHIVE_MAX_MB`). When a new archive write would push the
+  total over the cap, evict oldest entries (by `createdAt`) until we're
+  back under. Tool-name-agnostic, so it bounds memory regardless of which
+  tool returned the large payload.
+
+- [`8de45a7`](https://github.com/cesr/poncho-ai/commit/8de45a7ac434fa928ae3b83deec52727073d4658) Thanks [@cesr](https://github.com/cesr)! - fix(harness): browser status/frame listeners no longer pin runInput across runs
+
+  Heap-snapshot evidence pointed to the actual leak: `BrowserSession.tabs[cid].statusListeners`
+  was retaining ~3.4 GB on a long browser session. Each `harness.run()`
+  registered two arrow-function listeners (frame + status) whose lexical
+  scope captured the entire run scope, including `input.parameters.__toolResultArchive`.
+  V8 captures the full enclosing scope into the closure's Context object
+  even for variables the listener body doesn't reference, so the runInput
+  was reachable through every listener.
+
+  Two fixes:
+  1. The listeners are now produced by module-scope factories
+     (`makeBrowserFrameListener`, `makeBrowserStatusListener`) whose only
+     captured variable is the target event queue. The runInput is no longer
+     in scope when the closure is created.
+  2. The listener cleanup at the end of `run()` is now in a `try/finally`,
+     so listeners are always removed — even when the run errors or the
+     consumer abandons the generator. Previously a thrown run would leave
+     listeners pinned forever.
+
+- [`524df41`](https://github.com/cesr/poncho-ai/commit/524df411904bd00c07901695eda6d4dd07dde972) Thanks [@cesr](https://github.com/cesr)! - fix: persist harness messages on cancelled runs so the agent doesn't lose context
+
+  When a run was cancelled (Stop button, abort signal), `conversation.messages`
+  was updated with the partial assistant turn but `conversation._harnessMessages`
+  — the canonical history `loadCanonicalHistory` hands to the model on the next
+  turn — was left holding a snapshot from the _previous_ successful run. The
+  agent had no memory of the cancelled work, even though the user-facing UI
+  still showed it. The new verbose-mode harness toggle made this divergence
+  directly visible.
+
+  The fix plumbs an in-flight `messages` snapshot through the `run:cancelled`
+  event, trims it to a model-valid prefix (no orphan `tool_use`), and persists
+  it as `_harnessMessages` on every cancel path in the CLI.
+
+- [`8e410a1`](https://github.com/cesr/poncho-ai/commit/8e410a15b246a2b129fded8d1c06b98878e5fd07) Thanks [@cesr](https://github.com/cesr)! - fix(harness): surface real `isolated-vm` load error instead of generic message
+
+  The previous error told users "Run: pnpm add isolated-vm" even when the
+  package was installed but the native binary couldn't be loaded — typically
+  because a Node upgrade left the installed prebuilds with the wrong ABI
+  version (e.g. Node 25 reports ABI 141 but `isolated-vm@6.1.2` only ships
+  abi127/abi137 prebuilds). Now the error includes the underlying load
+  message, the current Node version + ABI, and a hint to rebuild rather
+  than reinstall when the cause is a binary mismatch.
+
+- [#105](https://github.com/cesr/poncho-ai/pull/105) [`e127174`](https://github.com/cesr/poncho-ai/commit/e12717415b1114c5e9a58e7c51fcf9e038218f9f) Thanks [@cesr](https://github.com/cesr)! - chore(harness): declare `engines.node` as `>=20.0.0 <25.0.0`
+
+  `isolated-vm@6.1.2` (the version harness uses for sandboxed code execution)
+  ships V8-ABI-specific prebuilt binaries up to ABI 137 (Node 24). Node 25
+  reports ABI 141 and has no matching prebuild, so the native module fails
+  to load. Declaring the upper bound makes pnpm/npm warn (or hard-fail with
+  `engine-strict`) at install time on Node 25, instead of surfacing as a
+  runtime error the first time `run_code` is invoked.
+
+- [`2792d84`](https://github.com/cesr/poncho-ai/commit/2792d8448b304bf748f926ce42a91c76f37edf79) Thanks [@cesr](https://github.com/cesr)! - Include weekday (Mon/Tue/...) alongside the UTC date in the system prompt's time context, so models stop misidentifying the day of the week.
+
+- Updated dependencies [[`524df41`](https://github.com/cesr/poncho-ai/commit/524df411904bd00c07901695eda6d4dd07dde972), [`9616060`](https://github.com/cesr/poncho-ai/commit/96160607502c2c0b05bc60b67b8fc012f4052ef1)]:
+  - @poncho-ai/sdk@1.10.0
+
+## 0.39.1
+
+### Patch Changes
+
+- [`244a3a3`](https://github.com/cesr/poncho-ai/commit/244a3a310c6c52f9e8535b28fb25d77829583d3f) Thanks [@cesr](https://github.com/cesr)! - fix(harness): don't archive `browser_screenshot` / `browser_snapshot` payloads
+
+  The per-conversation `_toolResultArchive` had no size cap or eviction, and
+  browser tool results were being archived in full — base64 JPEG screenshots
+  (~50-500KB each) and accessibility-tree snapshots accumulated for the lifetime
+  of a conversation. Heavy browser sessions OOM'd `poncho dev` after ~80 minutes.
+
+  Skip archiving for view-once tool results (`browser_screenshot`,
+  `browser_snapshot`). The model consumes them in-step; they're never retrieved
+  after-the-fact, so archiving them only burns memory.
+
+- [`d6248c8`](https://github.com/cesr/poncho-ai/commit/d6248c8b6d22e0fd0becde9e31dff7c12c724d84) Thanks [@cesr](https://github.com/cesr)! - fix(cli, harness): unify turn-parameter assembly so `conversation_recall` works everywhere
+
+  The recall tool relies on three context parameters (`__conversationRecallCorpus`,
+  `__conversationListFn`, `__conversationFetchFn`) that were only injected for
+  user-initiated HTTP turns. Cron, reminder, messaging-adapter, chat-continuation,
+  subagent-callback, and tool-approval-resume runs all built their own
+  `runInput.parameters` object and silently omitted these — causing
+  `conversation_recall` to throw "not available in this environment" or return
+  empty results depending on the call mode.
+
+  Introduces a single `buildTurnParameters(conversation, opts)` helper in the CLI
+  that owns context-parameter assembly (recall functions, `__activeConversationId`,
+  `__ownerId`, messaging metadata, tool-result archive). HTTP, messaging, and
+  cron/reminder paths now go through it. The harness orchestrator's three
+  internal turn sites (chat continuation, subagent-callback resume, tool-approval
+  resume) now call the existing `hooks.buildRecallParams` so they pick up the
+  recall functions too.
+
 ## 0.39.0
 
 ### Minor Changes

package/dist/index.d.ts

@@ -1030,7 +1030,6 @@ declare class AgentHarness {
     private readonly modelProviderInjected;
     private readonly dispatcher;
     readonly uploadStore?: UploadStore;
-    private skillContextWindow;
     private memoryStore?;
     private readonly tenantMemoryStores;
     private memoryConfig?;
@@ -1042,6 +1041,8 @@ declare class AgentHarness {
     private skillFingerprint;
     private lastSkillRefreshAt;
     private readonly activeSkillNames;
+    private readonly skillCache;
+    private readonly vfsSkillCollisionWarnings;
     private readonly registeredMcpToolNames;
     private otlpSpanProcessor?;
     private otlpTracerProvider?;
@@ -1085,6 +1086,15 @@ declare class AgentHarness {
      */
     private getMemoryStore;
     private listActiveSkills;
+    /**
+     * Resolve the skill set visible to a given tenant: repo skills plus that
+     * tenant's VFS skills, with repo winning on name collision. Cached per
+     * tenant; cache invalidates on VFS writes under /skills/ via
+     * invalidateSkillsForTenant.
+     */
+    private getSkillsForTenant;
+    invalidateSkillsForTenant(tenantId: string): void;
+    private computeVfsSkillFingerprint;
     private getAgentMcpIntent;
     private getAgentScriptIntent;
     private getAgentMcpApprovalPatterns;
@@ -1180,6 +1190,12 @@ declare const createSearchTools: () => ToolDefinition[];
  * Merges the defaults with any extra paths provided via config.
  */
 declare const resolveSkillDirs: (workingDir: string, extraPaths?: string[]) => string[];
+type SkillSource = {
+    kind: "repo";
+} | {
+    kind: "vfs";
+    tenantId: string;
+};
 interface SkillMetadata {
     /** Unique skill name from frontmatter. */
     name: string;
@@ -1194,9 +1210,11 @@ interface SkillMetadata {
         mcp: string[];
         scripts: string[];
     };
-    /** Absolute path to the skill directory. */
+    /** Where this skill came from. */
+    source: SkillSource;
+    /** Absolute fs path (repo) or VFS path (vfs) to the skill directory. */
     skillDir: string;
-    /** Absolute path to the SKILL.md file. */
+    /** Absolute fs path (repo) or VFS path (vfs) to the SKILL.md file. */
     skillPath: string;
 }
 /**
@@ -1205,29 +1223,39 @@ interface SkillMetadata {
 type SkillContextEntry = SkillMetadata & {
     instructions: string;
 };
+declare const parseSkillFrontmatter: (content: string) => {
+    name: string;
+    description: string;
+    allowedTools: {
+        mcp: string[];
+        scripts: string[];
+    };
+    approvalRequired: {
+        mcp: string[];
+        scripts: string[];
+    };
+} | undefined;
 declare const loadSkillMetadata: (workingDir: string, extraSkillPaths?: string[]) => Promise<SkillMetadata[]>;
 declare const buildSkillContextWindow: (skills: SkillMetadata[]) => string;
-declare const loadSkillInstructions: (skill: SkillMetadata) => Promise<string>;
-declare const readSkillResource: (skill: SkillMetadata, relativePath: string) => Promise<string>;
+declare const loadVfsSkillMetadata: (engine: StorageEngine, tenantId: string) => Promise<SkillMetadata[]>;
+declare const mergeSkills: (repoSkills: SkillMetadata[], vfsSkills: SkillMetadata[], onCollision?: (vfsSkill: SkillMetadata) => void) => SkillMetadata[];
+declare const loadSkillInstructions: (skill: SkillMetadata, engine?: StorageEngine) => Promise<string>;
+declare const readSkillResource: (skill: SkillMetadata, relativePath: string, engine?: StorageEngine) => Promise<string>;
 declare const loadSkillContext: (workingDir: string) => Promise<SkillContextEntry[]>;
 
-/**
- * Creates the built-in skill tools that implement progressive disclosure
- * per the Agent Skills specification (https://agentskills.io/integrate-skills).
- *
- * - `activate_skill`  — loads the full SKILL.md body on demand
- * - `read_skill_resource` — reads a file from a skill directory (references, scripts, assets)
- * - `list_skill_scripts` — lists runnable JavaScript/TypeScript scripts under scripts/
- * - `run_skill_script` — executes a JavaScript/TypeScript module under scripts/
- */
-declare const createSkillTools: (skills: SkillMetadata[], options?: {
+interface CreateSkillToolsOptions {
     workingDir?: string;
+    /** Resolve the skill set visible to a given tenant (repo + that tenant's VFS). */
+    getSkills: (tenantId: string | undefined | null) => Promise<SkillMetadata[]>;
+    /** Lazy accessor for the engine used for VFS reads when a skill's source is `vfs`. */
+    storageEngine?: () => StorageEngine | undefined;
     onActivateSkill?: (name: string) => Promise<string[]> | string[];
     onDeactivateSkill?: (name: string) => Promise<string[]> | string[];
     onListActiveSkills?: () => string[];
     isScriptAllowed?: (skill: string, scriptPath: string) => boolean;
     isRootScriptAllowed?: (scriptPath: string) => boolean;
-}) => ToolDefinition[];
+}
+declare const createSkillTools: (options: CreateSkillToolsOptions) => ToolDefinition[];
 declare const normalizeScriptPolicyPath: (relativePath: string) => string;
 
 declare const createReminderTools: (store: ReminderStore) => ToolDefinition[];
@@ -1810,4 +1838,4 @@ declare class AgentOrchestrator {
     recoverStaleSubagents(): Promise<void>;
 }
 
-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 CompactMessagesOptions, type CompactResult, type CompactionConfig, type ContinuationHooks, type Conversation, type ConversationCreateInit, type ConversationState, type ConversationStatusSnapshot, type ConversationStore, type ConversationSummary, type CronJobConfig, 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 NetworkConfig, 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 RunOutcome, type RunRequest, type RuntimeRenderContext, S3UploadStore, STALE_SUBAGENT_THRESHOLD_MS, STORAGE_SCHEMA_VERSION, type SecretsStore, type SkillContextEntry, type SkillMetadata, SqliteEngine, type StateConfig, type StateProviderName, type StateStore, type StorageConfig, type StorageEngine, type StorageFactoryOptions, type StorageProvider, type StoredApproval, type SubagentManager, type SubagentResult, type SubagentSpawnResult, type SubagentSummary, 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, 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, deleteOpenAICodexSession, deriveUploadKey, ensureAgentIdentity, estimateTokens, estimateTotalTokens, executeConversationTurn, findSafeSplitPoint, flushTurnDraft, generateAgentId, getAgentStoreDirectory, getModelContextWindow, getOpenAICodexAccessToken, getOpenAICodexAuthFilePath, getOpenAICodexRequiredScopes, getPonchoStoreRoot, isMessageArray, jsonSchemaToZod, loadCanonicalHistory, loadPonchoConfig, loadRunHistory, loadSkillContext, loadSkillInstructions, loadSkillMetadata, normalizeApprovalCheckpoint, normalizeOtlp, normalizeScriptPolicyPath, parseAgentFile, parseAgentMarkdown, ponchoDocsTool, readOpenAICodexSession, readSkillResource, recordStandardTurnEvent, renderAgentPrompt, resolveAgentIdentity, resolveCompactionConfig, resolveEnv, resolveMemoryConfig, resolveRunRequest, resolveSkillDirs, resolveStateConfig, slugifyStorageComponent, startOpenAICodexDeviceAuth, 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 CompactMessagesOptions, type CompactResult, type CompactionConfig, type ContinuationHooks, type Conversation, type ConversationCreateInit, type ConversationState, type ConversationStatusSnapshot, type ConversationStore, type ConversationSummary, type CreateSkillToolsOptions, type CronJobConfig, 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 NetworkConfig, 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 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 SubagentSpawnResult, type SubagentSummary, 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, 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, deleteOpenAICodexSession, deriveUploadKey, ensureAgentIdentity, estimateTokens, estimateTotalTokens, executeConversationTurn, findSafeSplitPoint, flushTurnDraft, generateAgentId, getAgentStoreDirectory, getModelContextWindow, getOpenAICodexAccessToken, getOpenAICodexAuthFilePath, getOpenAICodexRequiredScopes, getPonchoStoreRoot, isMessageArray, jsonSchemaToZod, loadCanonicalHistory, loadPonchoConfig, loadRunHistory, loadSkillContext, loadSkillInstructions, loadSkillMetadata, loadVfsSkillMetadata, mergeSkills, normalizeApprovalCheckpoint, normalizeOtlp, normalizeScriptPolicyPath, parseAgentFile, parseAgentMarkdown, parseSkillFrontmatter, ponchoDocsTool, readOpenAICodexSession, readSkillResource, recordStandardTurnEvent, renderAgentPrompt, resolveAgentIdentity, resolveCompactionConfig, resolveEnv, resolveMemoryConfig, resolveRunRequest, resolveSkillDirs, resolveStateConfig, slugifyStorageComponent, startOpenAICodexDeviceAuth, verifyTenantToken, withToolResultArchiveParam, writeOpenAICodexSession };