@poncho-ai/cli 0.40.2 → 0.40.4

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/cli@0.40.2 build /home/runner/work/poncho-ai/poncho-ai/packages/cli
+> @poncho-ai/cli@0.40.4 build /home/runner/work/poncho-ai/poncho-ai/packages/cli
 > tsup src/index.ts src/cli.ts --format esm --dts
 
 CLI Building entry: src/cli.ts, src/index.ts
@@ -9,10 +9,10 @@
 ESM Build start
 ESM dist/cli.js                          528.00 B
 ESM dist/index.js                        3.10 KB
-ESM dist/run-interactive-ink-LJTKUUV4.js 23.38 KB
-ESM dist/chunk-KVGMTYDD.js               674.92 KB
-ESM ⚡️ Build success in 83ms
+ESM dist/run-interactive-ink-3BKXJ5SX.js 23.38 KB
+ESM dist/chunk-4XGZI7KU.js               664.72 KB
+ESM ⚡️ Build success in 75ms
 DTS Build start
-DTS ⚡️ Build success in 4702ms
+DTS ⚡️ Build success in 4000ms
 DTS dist/cli.d.ts   20.00 B
-DTS dist/index.d.ts 13.25 KB
+DTS dist/index.d.ts 13.56 KB

package/CHANGELOG.md

@@ -1,5 +1,96 @@
 # @poncho-ai/cli
 
+## 0.40.4
+
+### Patch Changes
+
+- [`ff89631`](https://github.com/cesr/poncho-ai/commit/ff89631715e54d6fdce174943e6e0fc9e4ce5d1e) Thanks [@cesr](https://github.com/cesr)! - harness: export `defaultAgentDefinition` so SDK consumers can match `poncho init` exactly
+
+  Lifts the `AGENT_TEMPLATE` markdown body from `@poncho-ai/cli` (where it lived
+  inside the `init` scaffolding) into a public helper on `@poncho-ai/harness`.
+  SDK consumers (PonchOS, custom servers, anyone calling
+  `new AgentHarness({ agentDefinition })` directly) can now do:
+
+  ```ts
+  import { defaultAgentDefinition } from "@poncho-ai/harness";
+
+  const harness = new AgentHarness({
+    agentDefinition: defaultAgentDefinition({
+      name: "poncho",
+      modelName: "claude-sonnet-4-6",
+    }),
+    // ... storageEngine, config, etc.
+  });
+  ```
+
+  This eliminates hand-copying the template — drift between consumers and
+  `poncho init` is no longer possible.
+
+  The CLI's `AGENT_TEMPLATE` export is preserved as a thin back-compat
+  wrapper that delegates to `defaultAgentDefinition`. No behavior change.
+
+  API additions (harness):
+  - `defaultAgentDefinition(opts?: DefaultAgentDefinitionOptions): string`
+  - `DefaultAgentDefinitionOptions`
+  - `DEFAULT_AGENT_NAME`, `DEFAULT_AGENT_DESCRIPTION`,
+    `DEFAULT_MODEL_PROVIDER`, `DEFAULT_MODEL_NAME`, `DEFAULT_TEMPERATURE`,
+    `DEFAULT_MAX_STEPS`, `DEFAULT_TIMEOUT` constants
+
+- Updated dependencies [[`ff89631`](https://github.com/cesr/poncho-ai/commit/ff89631715e54d6fdce174943e6e0fc9e4ce5d1e)]:
+  - @poncho-ai/harness@0.43.0
+
+## 0.40.3
+
+### Patch Changes
+
+- [`111d24e`](https://github.com/cesr/poncho-ai/commit/111d24efaab054ef7543c396085f8f4d41e7976a) Thanks [@cesr](https://github.com/cesr)! - cli: include VFS skills in the chat input slash command menu
+
+  The `/api/slash-commands` endpoint was returning only repo-loaded skills,
+  so tenant-authored skills stored in the VFS (`/skills/<name>/SKILL.md`)
+  never appeared in the `/` autocomplete bar even though the agent could
+  already see and run them at conversation time.
+
+  The endpoint now resolves skills per-tenant via a new
+  `harness.listSkillsForTenant(tenantId)` and applies the same repo-wins
+  collision semantics used elsewhere in the harness.
+
+- [`39793b0`](https://github.com/cesr/poncho-ai/commit/39793b0ab11ed26f140af6fc9c0cd3e1b1c83fec) Thanks [@cesr](https://github.com/cesr)! - harness: extract `runConversationTurn` helper; refactor CLI to use it
+
+  Lifts the inline turn lifecycle from the CLI's
+  `POST /api/conversations/:id/messages` handler (~280 lines of orchestration)
+  into a new public helper at `@poncho-ai/harness`.
+
+  The helper handles the full conversation lifecycle for a primary chat
+  turn: load the conversation with archive, resolve canonical history,
+  upload files via the harness's upload store, build stable user/assistant
+  ids, persist the user message immediately, drive `executeConversationTurn`,
+  periodically persist the in-flight assistant draft on `step:completed`
+  and `tool:approval:required`, persist on `tool:approval:checkpoint` and
+  `run:completed` continuation, rebuild history on `compaction:completed`,
+  apply turn metadata on success, and persist partial state on
+  cancel/error.
+
+  Caller responsibilities (auth, active-run dedup, streaming, continuation
+  HTTP self-fetch, title inference) stay outside the helper — passed in
+  via opts or handled around the call. `opts.onEvent` is invoked for every
+  `AgentEvent` for downstream forwarding (SSE, WebSocket, telemetry, etc.).
+
+  The CLI's handler now delegates to `runConversationTurn` (drops from
+  ~430 to ~150 lines). Consumers like PonchOS can call the same helper
+  to ship the _exact_ same conversation lifecycle without duplicating
+  the orchestration.
+
+  Public API additions:
+  - `runConversationTurn(opts): Promise<RunConversationTurnResult>`
+  - `RunConversationTurnOpts`
+  - `RunConversationTurnResult`
+
+  No behavior changes. The helper is a verbatim extraction of the CLI's
+  prior inline implementation.
+
+- Updated dependencies [[`111d24e`](https://github.com/cesr/poncho-ai/commit/111d24efaab054ef7543c396085f8f4d41e7976a), [`39793b0`](https://github.com/cesr/poncho-ai/commit/39793b0ab11ed26f140af6fc9c0cd3e1b1c83fec)]:
+  - @poncho-ai/harness@0.42.0
+
 ## 0.40.2
 
 ### Patch Changes

package/dist/{chunk-KVGMTYDD.js → chunk-4XGZI7KU.js}

@@ -14,7 +14,6 @@ import {
   createConversationStore as createConversationStore2,
   createConversationStoreFromEngine as createConversationStoreFromEngine2,
   createUploadStore as createUploadStore2,
-  deriveUploadKey,
   ensureAgentIdentity as ensureAgentIdentity3,
   loadPonchoConfig as loadPonchoConfig4,
   parseAgentMarkdown as parseAgentMarkdown2,
@@ -32,6 +31,7 @@ import {
   normalizeApprovalCheckpoint,
   buildApprovalCheckpoints,
   applyTurnMetadata,
+  runConversationTurn,
   withToolResultArchiveParam,
   AgentOrchestrator
 } from "@poncho-ai/harness";
@@ -571,7 +571,7 @@ var WEB_UI_STYLES = `
     .sidebar-segmented {
       display: inline-flex;
       align-self: stretch;
-      margin: 12px 6px 0;
+      margin: 12px 6px 8px;
       padding: 3px;
       background: var(--surface-3);
       border-radius: 999px;
@@ -12028,6 +12028,7 @@ import { readFile as readFile4 } from "fs/promises";
 import { existsSync } from "fs";
 import { dirname as dirname4, relative, resolve as resolve3 } from "path";
 import { fileURLToPath } from "url";
+import { defaultAgentDefinition } from "@poncho-ai/harness";
 var __dirname = dirname4(fileURLToPath(import.meta.url));
 var packageRoot = resolve3(__dirname, "..");
 var readCliVersion = async () => {
@@ -12043,33 +12044,12 @@ var readCliVersion = async () => {
   }
   return fallback;
 };
-var AGENT_TEMPLATE = (name, id, options) => `---
-name: ${name}
-id: ${id}
-description: A helpful Poncho assistant
-model:
-  provider: ${options.modelProvider}
-  name: ${options.modelName}
-  temperature: 0.2
-limits:
-  maxSteps: 20
-  timeout: 300
----
-
-# {{name}}
-
-You are **{{name}}**, a helpful assistant built with Poncho.
-
-Working directory: {{runtime.workingDir}}
-Environment: {{runtime.environment}}
-
-## Task Guidance
-
-- Use tools when needed
-- Explain your reasoning clearly
-- Ask clarifying questions when requirements are ambiguous
-- Never claim a file/tool change unless the corresponding tool call actually succeeded
-`;
+var AGENT_TEMPLATE = (name, id, options) => defaultAgentDefinition({
+  name,
+  id,
+  modelProvider: options.modelProvider,
+  modelName: options.modelName
+});
 var resolveLocalPackagesRoot = () => {
   const candidate = resolve3(__dirname, "..", "..", "harness", "package.json");
   if (existsSync(candidate)) {
@@ -14261,7 +14241,7 @@ var runInteractive = async (workingDir, params) => {
   await harness.initialize();
   const identity = await ensureAgentIdentity2(workingDir);
   try {
-    const { runInteractiveInk } = await import("./run-interactive-ink-LJTKUUV4.js");
+    const { runInteractiveInk } = await import("./run-interactive-ink-3BKXJ5SX.js");
     await runInteractiveInk({
       harness,
       params,
@@ -14307,7 +14287,6 @@ var approvalLog = createLogger("approval");
 var browserLog = createLogger("browser");
 var selfFetchLog = createLogger("self-fetch");
 var csrfLog = createLogger("csrf");
-var uploadLog = createLogger("upload");
 var collectToolCallIds = (msgs) => {
   const ids = /* @__PURE__ */ new Set();
   for (const m of msgs) {
@@ -16826,7 +16805,8 @@ data: ${JSON.stringify(frame)}
       return;
     }
     if (pathname === "/api/slash-commands" && request.method === "GET") {
-      const skills = harness.listSkills().map((s) => ({
+      const tenantSkills = await harness.listSkillsForTenant(ctx.tenantId);
+      const skills = tenantSkills.map((s) => ({
         command: "/" + s.name,
         description: s.description,
         type: "skill"
@@ -16964,7 +16944,7 @@ data: ${JSON.stringify(frame)}
     const conversationMessageMatch = pathname.match(/^\/api\/conversations\/([^/]+)\/messages$/);
     if (conversationMessageMatch && request.method === "POST") {
       const conversationId = decodeURIComponent(conversationMessageMatch[1] ?? "");
-      const conversation = await conversationStore.getWithArchive(conversationId);
+      const conversation = await conversationStore.get(conversationId);
       if (!conversation || !canAccessConversation(conversation)) {
         writeJson(response, 404, {
           code: "CONVERSATION_NOT_FOUND",
@@ -17025,6 +17005,8 @@ data: ${JSON.stringify(frame)}
       });
       if (conversation.messages.length === 0 && (conversation.title === "New conversation" || conversation.title.trim().length === 0)) {
         conversation.title = inferConversationTitle(messageText);
+        conversation.updatedAt = Date.now();
+        await conversationStore.update(conversation);
       }
       response.writeHead(200, {
         "Content-Type": "text/event-stream",
@@ -17032,53 +17014,6 @@ data: ${JSON.stringify(frame)}
         Connection: "keep-alive",
         "X-Accel-Buffering": "no"
       });
-      const canonicalHistory = resolveRunRequest(conversation, {
-        conversationId,
-        messages: conversation.messages
-      });
-      const shouldRebuildCanonical = canonicalHistory.shouldRebuildCanonical;
-      const harnessMessages = [...canonicalHistory.messages];
-      const historyMessages = [...conversation.messages];
-      const preRunMessages = [...conversation.messages];
-      log.debug(
-        `conversation=${conversationId.slice(0, 8)} history=${canonicalHistory.source}`
-      );
-      let latestRunId = conversation.runtimeRunId ?? "";
-      let userContent = messageText;
-      if (files.length > 0) {
-        try {
-          const uploadedParts = await Promise.all(
-            files.map(async (f) => {
-              const buf = Buffer.from(f.data, "base64");
-              const key = deriveUploadKey(buf, f.mediaType);
-              const ref = await uploadStore.put(key, buf, f.mediaType);
-              return {
-                type: "file",
-                data: ref,
-                mediaType: f.mediaType,
-                filename: f.filename
-              };
-            })
-          );
-          userContent = [
-            { type: "text", text: messageText },
-            ...uploadedParts
-          ];
-        } catch (uploadErr) {
-          const errMsg = uploadErr instanceof Error ? uploadErr.message : String(uploadErr);
-          uploadLog.error(`file upload failed: ${errMsg}`);
-          const errorEvent = {
-            type: "run:error",
-            runId: "",
-            error: { code: "UPLOAD_ERROR", message: `File upload failed: ${errMsg}` }
-          };
-          broadcastEvent(conversationId, errorEvent);
-          finishConversationStream(conversationId);
-          activeConversationRuns.delete(conversationId);
-          response.end();
-          return;
-        }
-      }
       const unsubSubagentEvents = onConversationEvent(conversationId, (evt) => {
         if (evt.type.startsWith("subagent:")) {
           try {
@@ -17087,79 +17022,18 @@ data: ${JSON.stringify(frame)}
           }
         }
       });
-      const draft = createTurnDraftState();
-      let checkpointedRun = false;
-      let runCancelled = false;
-      let runContinuationMessages;
-      let cancelHarnessMessages;
-      const turnTimestamp = Date.now();
-      const userMessage = userContent != null ? {
-        role: "user",
-        content: userContent,
-        metadata: { id: randomUUID3(), timestamp: turnTimestamp }
-      } : void 0;
-      const assistantId = randomUUID3();
-      const buildMessages = () => {
-        const draftSections = cloneSections(draft.sections);
-        if (draft.currentTools.length > 0) {
-          draftSections.push({ type: "tools", content: [...draft.currentTools] });
-        }
-        if (draft.currentText.length > 0) {
-          draftSections.push({ type: "text", content: draft.currentText });
-        }
-        const userTurn = userMessage ? [userMessage] : [];
-        const hasDraftContent = draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draftSections.length > 0;
-        if (!hasDraftContent) {
-          return [...historyMessages, ...userTurn];
-        }
-        return [
-          ...historyMessages,
-          ...userTurn,
-          {
-            role: "assistant",
-            content: draft.assistantResponse,
-            metadata: buildAssistantMetadata2(draft, draftSections, { id: assistantId, timestamp: turnTimestamp })
-          }
-        ];
-      };
-      const persistDraftAssistantTurn = async () => {
-        if (draft.assistantResponse.length === 0 && draft.toolTimeline.length === 0) return;
-        conversation.messages = buildMessages();
-        conversation.updatedAt = Date.now();
-        await conversationStore.update(conversation);
-      };
+      let latestRunId = "";
       try {
-        {
-          conversation.messages = [
-            ...historyMessages,
-            ...userMessage ? [userMessage] : []
-          ];
-          conversation.subagentCallbackCount = 0;
-          conversation._continuationCount = void 0;
-          conversation.updatedAt = Date.now();
-          conversationStore.update(conversation).catch((err) => {
-            log.error(`failed to persist user turn: ${formatError(err)}`);
-          });
-        }
-        const execution = await executeConversationTurn2({
+        const result = await runConversationTurn({
           harness,
-          runInput: {
-            task: messageText,
-            conversationId,
-            tenantId: ctx.tenantId ?? void 0,
-            parameters: buildTurnParameters(conversation, { bodyParameters }),
-            messages: harnessMessages,
-            files: files.length > 0 ? files : void 0,
-            abortSignal: abortController.signal
-          },
-          initialContextTokens: conversation.contextTokens ?? 0,
-          initialContextWindow: conversation.contextWindow ?? 0,
-          onEvent: async (event, eventDraft) => {
-            draft.assistantResponse = eventDraft.assistantResponse;
-            draft.toolTimeline = eventDraft.toolTimeline;
-            draft.sections = eventDraft.sections;
-            draft.currentTools = eventDraft.currentTools;
-            draft.currentText = eventDraft.currentText;
+          conversationStore,
+          conversationId,
+          task: messageText,
+          files: files.length > 0 ? files : void 0,
+          parameters: buildTurnParameters(conversation, { bodyParameters }),
+          abortSignal: abortController.signal,
+          tenantId: ctx.tenantId ?? void 0,
+          onEvent: async (event) => {
             if (event.type === "run:started") {
               latestRunId = event.runId;
               runOwners.set(event.runId, ownerId);
@@ -17169,98 +17043,12 @@ data: ${JSON.stringify(frame)}
                 active.runId = event.runId;
               }
             }
-            if (event.type === "run:cancelled") {
-              runCancelled = true;
-              if (event.messages) cancelHarnessMessages = event.messages;
-            }
-            if (event.type === "compaction:completed") {
-              if (event.compactedMessages) {
-                historyMessages.length = 0;
-                historyMessages.push(...event.compactedMessages);
-                const preservedFromHistory = historyMessages.length - 1;
-                const removedCount = preRunMessages.length - Math.max(0, preservedFromHistory);
-                const existingHistory = conversation.compactedHistory ?? [];
-                conversation.compactedHistory = [
-                  ...existingHistory,
-                  ...preRunMessages.slice(0, removedCount)
-                ];
-              }
-            }
-            if (event.type === "step:completed") {
-              await persistDraftAssistantTurn();
-            }
-            if (event.type === "tool:approval:required") {
-              const toolText = `- approval required \`${event.tool}\``;
-              draft.toolTimeline.push(toolText);
-              draft.currentTools.push(toolText);
-              const existingApprovals = Array.isArray(conversation.pendingApprovals) ? conversation.pendingApprovals : [];
-              if (!existingApprovals.some((approval) => approval.approvalId === event.approvalId)) {
-                conversation.pendingApprovals = [
-                  ...existingApprovals,
-                  {
-                    approvalId: event.approvalId,
-                    runId: latestRunId || conversation.runtimeRunId || "",
-                    tool: event.tool,
-                    toolCallId: void 0,
-                    input: event.input ?? {},
-                    checkpointMessages: void 0,
-                    baseMessageCount: historyMessages.length,
-                    pendingToolCalls: []
-                  }
-                ];
-                conversation.updatedAt = Date.now();
-                await conversationStore.update(conversation);
-              }
-              await persistDraftAssistantTurn();
-            }
-            if (event.type === "tool:approval:checkpoint") {
-              conversation.messages = buildMessages();
-              conversation.pendingApprovals = buildApprovalCheckpoints({
-                approvals: event.approvals,
-                runId: latestRunId,
-                checkpointMessages: event.checkpointMessages,
-                baseMessageCount: historyMessages.length,
-                pendingToolCalls: event.pendingToolCalls
-              });
-              conversation._toolResultArchive = harness.getToolResultArchive(conversationId);
-              conversation.updatedAt = Date.now();
-              await conversationStore.update(conversation);
-              checkpointedRun = true;
-            }
-            if (event.type === "run:completed") {
-              if (event.result.continuation && event.result.continuationMessages) {
-                runContinuationMessages = event.result.continuationMessages;
-                conversation.messages = buildMessages();
-                conversation._continuationMessages = runContinuationMessages;
-                conversation._harnessMessages = runContinuationMessages;
-                conversation._toolResultArchive = harness.getToolResultArchive(conversationId);
-                conversation.runtimeRunId = latestRunId || conversation.runtimeRunId;
-                if (!checkpointedRun) {
-                  conversation.pendingApprovals = [];
-                }
-                if ((event.result.contextTokens ?? 0) > 0) conversation.contextTokens = event.result.contextTokens;
-                if ((event.result.contextWindow ?? 0) > 0) conversation.contextWindow = event.result.contextWindow;
-                conversation.updatedAt = Date.now();
-                await conversationStore.update(conversation);
-                if (!checkpointedRun) {
-                  doWaitUntil(
-                    new Promise((r) => setTimeout(r, 3e3)).then(
-                      () => selfFetchWithRetry(`/api/internal/continue/${encodeURIComponent(conversationId)}`)
-                    )
-                  );
-                }
-              }
-            }
             await telemetry.emit(event);
             let sseEvent = event.type === "compaction:completed" && event.compactedMessages ? { ...event, compactedMessages: void 0 } : event;
             if (sseEvent.type === "run:completed") {
               const hasPendingSubagents = await hasPendingSubagentWorkForParent(conversationId, ownerId);
               const stripped = { ...sseEvent, result: { ...sseEvent.result, continuationMessages: void 0 } };
-              if (hasPendingSubagents) {
-                sseEvent = { ...stripped, pendingSubagents: true };
-              } else {
-                sseEvent = stripped;
-              }
+              sseEvent = hasPendingSubagents ? { ...stripped, pendingSubagents: true } : stripped;
             }
             broadcastEvent(conversationId, sseEvent);
             try {
@@ -17270,38 +17058,15 @@ data: ${JSON.stringify(frame)}
             emitBrowserStatusIfActive(conversationId, event, response);
           }
         });
-        flushTurnDraft2(draft);
-        latestRunId = execution.latestRunId || latestRunId;
-        if (!checkpointedRun && !runContinuationMessages) {
-          conversation.messages = buildMessages();
-          applyTurnMetadata(conversation, {
-            latestRunId,
-            contextTokens: execution.runContextTokens,
-            contextWindow: execution.runContextWindow,
-            harnessMessages: execution.runHarnessMessages,
-            toolResultArchive: harness.getToolResultArchive(conversationId)
-          }, { shouldRebuildCanonical });
-          await conversationStore.update(conversation);
+        if (result.continuation && !result.checkpointed) {
+          doWaitUntil(
+            new Promise((r) => setTimeout(r, 3e3)).then(
+              () => selfFetchWithRetry(`/api/internal/continue/${encodeURIComponent(conversationId)}`)
+            )
+          );
         }
       } catch (error) {
-        flushTurnDraft2(draft);
-        if (abortController.signal.aborted || runCancelled) {
-          if (draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draft.sections.length > 0) {
-            conversation.messages = buildMessages();
-            applyTurnMetadata(conversation, {
-              latestRunId,
-              contextTokens: 0,
-              contextWindow: 0,
-              harnessMessages: cancelHarnessMessages,
-              toolResultArchive: harness.getToolResultArchive(conversationId)
-            }, { shouldRebuildCanonical: true });
-            await conversationStore.update(conversation);
-          }
-          if (!checkpointedRun) {
-            await clearPendingApprovalsForConversation(conversationId);
-          }
-          return;
-        }
+        log.error(`runConversationTurn threw: ${formatError(error)}`);
         try {
           response.write(
             formatSseEvent({
@@ -17314,11 +17079,6 @@ data: ${JSON.stringify(frame)}
             })
           );
         } catch {
-          if (draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draft.sections.length > 0) {
-            conversation.messages = buildMessages();
-            conversation.updatedAt = Date.now();
-            await conversationStore.update(conversation);
-          }
         }
       } finally {
         unsubSubagentEvents();

package/dist/cli.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   main
-} from "./chunk-KVGMTYDD.js";
+} from "./chunk-4XGZI7KU.js";
 
 // src/cli.ts
 var _originalEmitWarning = process.emitWarning.bind(process);

package/dist/index.d.ts

@@ -54,6 +54,13 @@ declare const MAX_PRUNE_PER_RUN = 25;
 /** Delete old cron conversations beyond `maxRuns`, capped to avoid API storms on catch-up. */
 declare const pruneCronConversations: (store: ConversationStore, ownerId: string, jobName: string, maxRuns: number) => Promise<number>;
 
+/**
+ * Thin back-compat wrapper around `defaultAgentDefinition` from the harness
+ * package. The canonical template now lives in
+ * `@poncho-ai/harness/src/default-agent.ts` so SDK consumers can pass the
+ * exact same default to `new AgentHarness({ agentDefinition })` without
+ * hand-copying the template.
+ */
 declare const AGENT_TEMPLATE: (name: string, id: string, options: {
     modelProvider: "anthropic" | "openai" | "openai-codex";
     modelName: string;

package/dist/index.js

@@ -77,7 +77,7 @@ import {
   writeHtml,
   writeJson,
   writeScaffoldFile
-} from "./chunk-KVGMTYDD.js";
+} from "./chunk-4XGZI7KU.js";
 export {
   AGENT_TEMPLATE,
   ENV_TEMPLATE,

package/dist/{run-interactive-ink-LJTKUUV4.js → run-interactive-ink-3BKXJ5SX.js}

@@ -3,7 +3,7 @@ import {
   getMascotLines,
   inferConversationTitle,
   resolveHarnessEnvironment
-} from "./chunk-KVGMTYDD.js";
+} from "./chunk-4XGZI7KU.js";
 
 // src/run-interactive-ink.ts
 import * as readline from "readline";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/cli",
-  "version": "0.40.2",
+  "version": "0.40.4",
   "description": "CLI for building and deploying AI agents",
   "repository": {
     "type": "git",
@@ -28,8 +28,8 @@
     "react": "^19.2.4",
     "react-devtools-core": "^6.1.5",
     "yaml": "^2.8.1",
-    "@poncho-ai/harness": "0.41.0",
     "@poncho-ai/messaging": "0.8.5",
+    "@poncho-ai/harness": "0.43.0",
     "@poncho-ai/sdk": "1.10.0"
   },
   "devDependencies": {

package/src/index.ts

@@ -16,7 +16,6 @@ import {
   createConversationStore,
   createConversationStoreFromEngine,
   createUploadStore,
-  deriveUploadKey,
   ensureAgentIdentity,
   loadPonchoConfig,
   parseAgentMarkdown,
@@ -39,6 +38,7 @@ import {
   normalizeApprovalCheckpoint,
   buildApprovalCheckpoints,
   applyTurnMetadata,
+  runConversationTurn,
   TOOL_RESULT_ARCHIVE_PARAM,
   withToolResultArchiveParam,
   AgentOrchestrator,
@@ -93,7 +93,6 @@ const approvalLog = createLogger("approval");
 const browserLog = createLogger("browser");
 const selfFetchLog = createLogger("self-fetch");
 const csrfLog = createLogger("csrf");
-const uploadLog = createLogger("upload");
 
 /**
  * Walk a sequence of harness messages and collect all tool-call ids that
@@ -3209,7 +3208,8 @@ export const createRequestHandler = async (options?: {
     }
 
     if (pathname === "/api/slash-commands" && request.method === "GET") {
-      const skills: ApiSlashCommand[] = harness.listSkills().map((s) => ({
+      const tenantSkills = await harness.listSkillsForTenant(ctx.tenantId);
+      const skills: ApiSlashCommand[] = tenantSkills.map((s) => ({
         command: "/" + s.name,
         description: s.description,
         type: "skill" as const,
@@ -3353,9 +3353,9 @@ export const createRequestHandler = async (options?: {
     const conversationMessageMatch = pathname.match(/^\/api\/conversations\/([^/]+)\/messages$/);
     if (conversationMessageMatch && request.method === "POST") {
       const conversationId = decodeURIComponent(conversationMessageMatch[1] ?? "");
-      // getWithArchive — conversation feeds withToolResultArchiveParam when
-      // the turn below calls executeConversationTurn.
-      const conversation = await conversationStore.getWithArchive(conversationId);
+      // Light access check first; the helper reloads the conversation
+      // (with archive) when it actually runs the turn.
+      const conversation = await conversationStore.get(conversationId);
       if (!conversation || !canAccessConversation(conversation)) {
         writeJson(response, 404, {
           code: "CONVERSATION_NOT_FOUND",
@@ -3420,159 +3420,46 @@ export const createRequestHandler = async (options?: {
         abortController,
         runId: null,
       });
+
+      // Auto-infer a title for fresh conversations. Persist it before the
+      // helper reloads, so its in-memory copy carries the new title.
       if (
         conversation.messages.length === 0 &&
         (conversation.title === "New conversation" || conversation.title.trim().length === 0)
       ) {
         conversation.title = inferConversationTitle(messageText);
+        conversation.updatedAt = Date.now();
+        await conversationStore.update(conversation);
       }
+
       response.writeHead(200, {
         "Content-Type": "text/event-stream",
         "Cache-Control": "no-cache",
         Connection: "keep-alive",
         "X-Accel-Buffering": "no",
       });
-      const canonicalHistory = resolveRunRequest(conversation, {
-        conversationId,
-        messages: conversation.messages,
-      });
-      const shouldRebuildCanonical = canonicalHistory.shouldRebuildCanonical;
-      const harnessMessages = [...canonicalHistory.messages];
-      const historyMessages = [...conversation.messages];
-      const preRunMessages = [...conversation.messages];
-      log.debug(
-        `conversation=${conversationId.slice(0, 8)} history=${canonicalHistory.source}`,
-      );
-      let latestRunId = conversation.runtimeRunId ?? "";
-      let userContent: Message["content"] | undefined = messageText;
-      if (files.length > 0) {
-        try {
-          const uploadedParts = await Promise.all(
-            files.map(async (f) => {
-              const buf = Buffer.from(f.data, "base64");
-              const key = deriveUploadKey(buf, f.mediaType);
-              const ref = await uploadStore.put(key, buf, f.mediaType);
-              return {
-                type: "file" as const,
-                data: ref,
-                mediaType: f.mediaType,
-                filename: f.filename,
-              };
-            }),
-          );
-          userContent = [
-            { type: "text" as const, text: messageText },
-            ...uploadedParts,
-          ];
-        } catch (uploadErr) {
-          const errMsg = uploadErr instanceof Error ? uploadErr.message : String(uploadErr);
-          uploadLog.error(`file upload failed: ${errMsg}`);
-          const errorEvent: AgentEvent = {
-            type: "run:error",
-            runId: "",
-            error: { code: "UPLOAD_ERROR", message: `File upload failed: ${errMsg}` },
-          };
-          broadcastEvent(conversationId, errorEvent);
-          finishConversationStream(conversationId);
-          activeConversationRuns.delete(conversationId);
-          response.end();
-          return;
-        }
-      }
+
+      // Forward subagent events out-of-band — they're emitted from
+      // child-conversation runs, not yielded by harness.run for this one.
       const unsubSubagentEvents = onConversationEvent(conversationId, (evt) => {
         if (evt.type.startsWith("subagent:")) {
           try { response.write(formatSseEvent(evt)); } catch {}
         }
       });
 
-      const draft = createTurnDraftState();
-      let checkpointedRun = false;
-      let runCancelled = false;
-      let runContinuationMessages: Message[] | undefined;
-      // Snapshot of the harness's in-flight messages emitted with run:cancelled,
-      // so the catch-path (executeConversationTurn threw) can still persist a
-      // canonical history that includes the cancelled work.
-      let cancelHarnessMessages: Message[] | undefined;
-
-      // Hoist stable ids for this turn. The same userMessage / assistantId is
-      // reused across every buildMessages() call so the in-flight assistant
-      // bubble keeps a stable metadata.id from its very first persisted byte.
-      const turnTimestamp = Date.now();
-      const userMessage: Message | undefined = userContent != null
-        ? {
-            role: "user" as const,
-            content: userContent,
-            metadata: { id: randomUUID(), timestamp: turnTimestamp },
-          }
-        : undefined;
-      const assistantId = randomUUID();
-
-      const buildMessages = (): Message[] => {
-        const draftSections = cloneSections(draft.sections);
-        if (draft.currentTools.length > 0) {
-          draftSections.push({ type: "tools", content: [...draft.currentTools] });
-        }
-        if (draft.currentText.length > 0) {
-          draftSections.push({ type: "text", content: draft.currentText });
-        }
-        const userTurn: Message[] = userMessage ? [userMessage] : [];
-        const hasDraftContent =
-          draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draftSections.length > 0;
-        if (!hasDraftContent) {
-          return [...historyMessages, ...userTurn];
-        }
-        return [
-          ...historyMessages,
-          ...userTurn,
-          {
-            role: "assistant" as const,
-            content: draft.assistantResponse,
-            metadata: buildAssistantMetadata(draft, draftSections, { id: assistantId, timestamp: turnTimestamp }),
-          },
-        ];
-      };
-
-      const persistDraftAssistantTurn = async (): Promise<void> => {
-        if (draft.assistantResponse.length === 0 && draft.toolTimeline.length === 0) return;
-        conversation.messages = buildMessages();
-        conversation.updatedAt = Date.now();
-        await conversationStore.update(conversation);
-      };
+      let latestRunId = "";
 
       try {
-        {
-          conversation.messages = [
-            ...historyMessages,
-            ...(userMessage ? [userMessage] : []),
-          ];
-          conversation.subagentCallbackCount = 0;
-          conversation._continuationCount = undefined;
-          conversation.updatedAt = Date.now();
-          conversationStore.update(conversation).catch((err) => {
-            log.error(`failed to persist user turn: ${formatError(err)}`);
-          });
-        }
-
-        const execution = await executeConversationTurn({
+        const result = await runConversationTurn({
           harness,
-          runInput: {
-            task: messageText,
-            conversationId,
-            tenantId: ctx.tenantId ?? undefined,
-            parameters: buildTurnParameters(conversation, { bodyParameters }),
-            messages: harnessMessages,
-            files: files.length > 0 ? files : undefined,
-            abortSignal: abortController.signal,
-          },
-          initialContextTokens: conversation.contextTokens ?? 0,
-          initialContextWindow: conversation.contextWindow ?? 0,
-          onEvent: async (event, eventDraft) => {
-            draft.assistantResponse = eventDraft.assistantResponse;
-            draft.toolTimeline = eventDraft.toolTimeline;
-            draft.sections = eventDraft.sections;
-            draft.currentTools = eventDraft.currentTools;
-            draft.currentText = eventDraft.currentText;
-
+          conversationStore,
+          conversationId,
+          task: messageText,
+          files: files.length > 0 ? files : undefined,
+          parameters: buildTurnParameters(conversation, { bodyParameters }),
+          abortSignal: abortController.signal,
+          tenantId: ctx.tenantId ?? undefined,
+          onEvent: async (event) => {
             if (event.type === "run:started") {
               latestRunId = event.runId;
               runOwners.set(event.runId, ownerId);
@@ -3582,154 +3469,36 @@ export const createRequestHandler = async (options?: {
                 active.runId = event.runId;
               }
             }
-            if (event.type === "run:cancelled") {
-              runCancelled = true;
-              if (event.messages) cancelHarnessMessages = event.messages;
-            }
-            if (event.type === "compaction:completed") {
-              if (event.compactedMessages) {
-                historyMessages.length = 0;
-                historyMessages.push(...event.compactedMessages);
-
-                const preservedFromHistory = historyMessages.length - 1;
-                const removedCount = preRunMessages.length - Math.max(0, preservedFromHistory);
-                const existingHistory = conversation.compactedHistory ?? [];
-                conversation.compactedHistory = [
-                  ...existingHistory,
-                  ...preRunMessages.slice(0, removedCount),
-                ];
-              }
-            }
-            if (event.type === "step:completed") {
-              await persistDraftAssistantTurn();
-            }
-            if (event.type === "tool:approval:required") {
-              const toolText = `- approval required \`${event.tool}\``;
-              draft.toolTimeline.push(toolText);
-              draft.currentTools.push(toolText);
-              const existingApprovals = Array.isArray(conversation.pendingApprovals)
-                ? conversation.pendingApprovals
-                : [];
-              if (!existingApprovals.some((approval) => approval.approvalId === event.approvalId)) {
-                conversation.pendingApprovals = [
-                  ...existingApprovals,
-                  {
-                    approvalId: event.approvalId,
-                    runId: latestRunId || conversation.runtimeRunId || "",
-                    tool: event.tool,
-                    toolCallId: undefined,
-                    input: (event.input ?? {}) as Record<string, unknown>,
-                    checkpointMessages: undefined,
-                    baseMessageCount: historyMessages.length,
-                    pendingToolCalls: [],
-                  },
-                ];
-                conversation.updatedAt = Date.now();
-                await conversationStore.update(conversation);
-              }
-              await persistDraftAssistantTurn();
-            }
-            if (event.type === "tool:approval:checkpoint") {
-              conversation.messages = buildMessages();
-              conversation.pendingApprovals = buildApprovalCheckpoints({
-                approvals: event.approvals,
-                runId: latestRunId,
-                checkpointMessages: event.checkpointMessages,
-                baseMessageCount: historyMessages.length,
-                pendingToolCalls: event.pendingToolCalls,
-              });
-              conversation._toolResultArchive = harness.getToolResultArchive(conversationId);
-              conversation.updatedAt = Date.now();
-              await conversationStore.update(conversation);
-              checkpointedRun = true;
-            }
-            if (event.type === "run:completed") {
-              if (event.result.continuation && event.result.continuationMessages) {
-                runContinuationMessages = event.result.continuationMessages;
-
-                conversation.messages = buildMessages();
-                conversation._continuationMessages = runContinuationMessages;
-                conversation._harnessMessages = runContinuationMessages;
-                conversation._toolResultArchive = harness.getToolResultArchive(conversationId);
-                conversation.runtimeRunId = latestRunId || conversation.runtimeRunId;
-                if (!checkpointedRun) {
-                  conversation.pendingApprovals = [];
-                }
-                if ((event.result.contextTokens ?? 0) > 0) conversation.contextTokens = event.result.contextTokens!;
-                if ((event.result.contextWindow ?? 0) > 0) conversation.contextWindow = event.result.contextWindow!;
-                conversation.updatedAt = Date.now();
-                await conversationStore.update(conversation);
-
-                if (!checkpointedRun) {
-                  doWaitUntil(
-                    new Promise(r => setTimeout(r, 3000)).then(() =>
-                      selfFetchWithRetry(`/api/internal/continue/${encodeURIComponent(conversationId)}`),
-                    ),
-                  );
-                }
-              }
-            }
-
             await telemetry.emit(event);
+            // Strip large internal payloads before sending over the wire.
             let sseEvent: AgentEvent = event.type === "compaction:completed" && event.compactedMessages
               ? { ...event, compactedMessages: undefined }
               : event;
             if (sseEvent.type === "run:completed") {
               const hasPendingSubagents = await hasPendingSubagentWorkForParent(conversationId, ownerId);
               const stripped = { ...sseEvent, result: { ...sseEvent.result, continuationMessages: undefined } };
-              if (hasPendingSubagents) {
-                sseEvent = { ...stripped, pendingSubagents: true };
-              } else {
-                sseEvent = stripped;
-              }
+              sseEvent = hasPendingSubagents ? { ...stripped, pendingSubagents: true } : stripped;
             }
             broadcastEvent(conversationId, sseEvent);
-            try {
-              response.write(formatSseEvent(sseEvent));
-            } catch {
-              // Client disconnected — continue processing so the run completes.
-            }
+            try { response.write(formatSseEvent(sseEvent)); } catch {}
             emitBrowserStatusIfActive(conversationId, event, response);
           },
         });
 
-        flushTurnDraft(draft);
-        latestRunId = execution.latestRunId || latestRunId;
-
-        if (!checkpointedRun && !runContinuationMessages) {
-          conversation.messages = buildMessages();
-          applyTurnMetadata(conversation, {
-            latestRunId,
-            contextTokens: execution.runContextTokens,
-            contextWindow: execution.runContextWindow,
-            harnessMessages: execution.runHarnessMessages,
-            toolResultArchive: harness.getToolResultArchive(conversationId),
-          }, { shouldRebuildCanonical });
-          await conversationStore.update(conversation);
+        // Trigger the auto-continuation HTTP roundtrip so the next run
+        // queues. Helper has already persisted the continuation messages.
+        if (result.continuation && !result.checkpointed) {
+          doWaitUntil(
+            new Promise(r => setTimeout(r, 3000)).then(() =>
+              selfFetchWithRetry(`/api/internal/continue/${encodeURIComponent(conversationId)}`),
+            ),
+          );
         }
       } catch (error) {
-        flushTurnDraft(draft);
-        if (abortController.signal.aborted || runCancelled) {
-          if (draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draft.sections.length > 0) {
-            conversation.messages = buildMessages();
-            // Keep _harnessMessages aligned with what the model just saw.
-            // Without this, loadCanonicalHistory will hand the next turn a
-            // pre-cancellation snapshot and the agent will have no memory of
-            // the work it just did.
-            applyTurnMetadata(conversation, {
-              latestRunId,
-              contextTokens: 0,
-              contextWindow: 0,
-              harnessMessages: cancelHarnessMessages,
-              toolResultArchive: harness.getToolResultArchive(conversationId),
-            }, { shouldRebuildCanonical: true });
-            await conversationStore.update(conversation);
-          }
-          if (!checkpointedRun) {
-            await clearPendingApprovalsForConversation(conversationId);
-          }
-          return;
-        }
+        // The helper handles cancel/error internally and returns gracefully.
+        // This catches anything that escapes — typically a helper bug or
+        // the conversation being deleted between auth check and helper reload.
+        log.error(`runConversationTurn threw: ${formatError(error)}`);
         try {
           response.write(
             formatSseEvent({
@@ -3741,13 +3510,7 @@ export const createRequestHandler = async (options?: {
               },
             }),
           );
-        } catch {
-          if (draft.assistantResponse.length > 0 || draft.toolTimeline.length > 0 || draft.sections.length > 0) {
-            conversation.messages = buildMessages();
-            conversation.updatedAt = Date.now();
-            await conversationStore.update(conversation);
-          }
-        }
+        } catch {}
       } finally {
         unsubSubagentEvents();
         const active = activeConversationRuns.get(conversationId);

package/src/templates.ts

@@ -2,6 +2,7 @@ import { readFile } from "node:fs/promises";
 import { existsSync } from "node:fs";
 import { dirname, relative, resolve } from "node:path";
 import { fileURLToPath } from "node:url";
+import { defaultAgentDefinition } from "@poncho-ai/harness";
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const packageRoot = resolve(__dirname, "..");
@@ -21,37 +22,24 @@ const readCliVersion = async (): Promise<string> => {
   return fallback;
 };
 
+/**
+ * Thin back-compat wrapper around `defaultAgentDefinition` from the harness
+ * package. The canonical template now lives in
+ * `@poncho-ai/harness/src/default-agent.ts` so SDK consumers can pass the
+ * exact same default to `new AgentHarness({ agentDefinition })` without
+ * hand-copying the template.
+ */
 export const AGENT_TEMPLATE = (
   name: string,
   id: string,
   options: { modelProvider: "anthropic" | "openai" | "openai-codex"; modelName: string },
-): string => `---
-name: ${name}
-id: ${id}
-description: A helpful Poncho assistant
-model:
-  provider: ${options.modelProvider}
-  name: ${options.modelName}
-  temperature: 0.2
-limits:
-  maxSteps: 20
-  timeout: 300
----
-
-# {{name}}
-
-You are **{{name}}**, a helpful assistant built with Poncho.
-
-Working directory: {{runtime.workingDir}}
-Environment: {{runtime.environment}}
-
-## Task Guidance
-
-- Use tools when needed
-- Explain your reasoning clearly
-- Ask clarifying questions when requirements are ambiguous
-- Never claim a file/tool change unless the corresponding tool call actually succeeded
-`;
+): string =>
+  defaultAgentDefinition({
+    name,
+    id,
+    modelProvider: options.modelProvider,
+    modelName: options.modelName,
+  });
 
 /**
  * Resolve the monorepo packages root if we're running from a local dev build.

package/src/web-ui-styles.ts

@@ -439,7 +439,7 @@ export const WEB_UI_STYLES = `
     .sidebar-segmented {
       display: inline-flex;
       align-self: stretch;
-      margin: 12px 6px 0;
+      margin: 12px 6px 8px;
       padding: 3px;
       background: var(--surface-3);
       border-radius: 999px;