@poncho-ai/harness 0.60.1 → 0.61.1

package/dist/index.js

@@ -232,9 +232,10 @@ var parseAgentMarkdown = (content) => {
           "Invalid AGENT.md frontmatter compaction.trigger: must be between 0.1 and 1."
         );
       }
+      const keepRecentTurns = typeof raw.keepRecentTurns === "number" ? Math.max(1, Math.floor(raw.keepRecentTurns)) : 4;
       const keepRecentMessages = typeof raw.keepRecentMessages === "number" ? Math.max(2, Math.floor(raw.keepRecentMessages)) : 6;
       const instructions = typeof raw.instructions === "string" && raw.instructions.trim() ? raw.instructions.trim() : void 0;
-      return { enabled, trigger, keepRecentMessages, instructions };
+      return { enabled, trigger, keepRecentTurns, keepRecentMessages, instructions };
     })(),
     cron: parseCronJobs(parsed.cron)
   };
@@ -359,29 +360,37 @@ var MIN_COMPACTABLE_MESSAGES = 4;
 var DEFAULT_COMPACTION_CONFIG = {
   enabled: true,
   trigger: 0.75,
-  keepRecentMessages: 4
+  keepRecentTurns: 4,
+  keepRecentMessages: 6
 };
-var SUMMARIZATION_MESSAGE_TRUNCATION_CHARS = 1200;
-var SUMMARIZATION_MAX_OUTPUT_TOKENS = 768;
+var SUMMARIZATION_MESSAGE_TRUNCATION_CHARS = 4e3;
+var SUMMARIZATION_MAX_OUTPUT_TOKENS = 8192;
+var SUMMARIZATION_MAX_INPUT_TOKENS = 12e4;
 var SUMMARIZATION_PROMPT = `Summarize the following conversation into a structured working state that allows continuation without re-asking questions. Include:
 
 1. **User intent**: What the user originally asked for and any refinements
-2. **Completed work**: What has been accomplished so far
+2. **Completed work**: What has been accomplished AND CONFIRMED so far
 3. **Key decisions**: Technical decisions made and their rationale
-4. **Errors & fixes**: Errors encountered and how they were resolved
-5. **Referenced resources**: Files, URLs, tools, or data referenced
-6. **Pending next steps**: What remains to be done
+4. **Unresolved errors & failures**: Every error, failed or uncertain tool call, and failed or incomplete subagent \u2014 recorded verbatim enough to act on. NEVER drop a failure because it is old. This section is REQUIRED whenever any failure occurred; do not omit it.
+5. **Pending promises**: Anything the assistant said it WOULD do but has not confirmed done.
+6. **Referenced resources**: Files, URLs, tools, or data referenced
+7. **Pending next steps**: What remains to be done
+
+Rules:
+- NEVER describe work as "completed", "done", or "fixed" unless the conversation contains explicit confirmation (a successful tool result or the user acknowledging it). If completion is unverified, record it under "Pending next steps" as unconfirmed.
+- Preserve identifiers verbatim \u2014 file paths, URLs, IDs, command names, tool names. Do not paraphrase or truncate them.
 
 Be concise but preserve all information needed to continue the task.
-Omit any section that has no relevant content.`;
+Omit any section that has no relevant content, EXCEPT "Unresolved errors & failures" when a failure occurred.`;
 var CUMULATIVE_SUMMARY_PROMPT = `The FIRST message below (tagged [prior-summary]) is an existing working-state summary produced by an earlier compaction. Treat it as the authoritative prior working state: MERGE AND UPDATE it with the newer messages that follow it, carrying forward all still-relevant detail. Do NOT discard or re-compress information from the prior summary just because it is older \u2014 only drop it if the newer messages explicitly supersede it.`;
-var SUBAGENT_DIGEST_CHARS = 500;
+var SUBAGENT_DIGEST_CHARS = 2e3;
 var SUBAGENT_LEDGER_HEADING = "## Subagents";
 var resolveCompactionConfig = (explicit) => {
   if (!explicit) return { ...DEFAULT_COMPACTION_CONFIG };
   return {
     enabled: explicit.enabled ?? DEFAULT_COMPACTION_CONFIG.enabled,
     trigger: explicit.trigger ?? DEFAULT_COMPACTION_CONFIG.trigger,
+    keepRecentTurns: explicit.keepRecentTurns ?? DEFAULT_COMPACTION_CONFIG.keepRecentTurns,
     keepRecentMessages: explicit.keepRecentMessages ?? DEFAULT_COMPACTION_CONFIG.keepRecentMessages,
     instructions: explicit.instructions
   };
@@ -433,18 +442,57 @@ var findSafeSplitPoint = (messages, keepRecentMessages) => {
   }
   return -1;
 };
+var findSafeSplitPointByTurns = (messages, keepRecentTurns, keepRecentMessagesFallback, maxPreservedTokens) => {
+  const userIdx = [];
+  for (let i = 0; i < messages.length; i++) {
+    if (messages[i].role === "user") userIdx.push(i);
+  }
+  const maxN = Math.min(keepRecentTurns, userIdx.length);
+  for (let n = maxN; n >= 1; n--) {
+    let guardN = n;
+    let split = userIdx[userIdx.length - guardN];
+    while (splitOrphansToolCalls(messages, split) && guardN < userIdx.length) {
+      guardN++;
+      split = userIdx[userIdx.length - guardN];
+    }
+    if (split < MIN_COMPACTABLE_MESSAGES) continue;
+    const preservedTokens = estimateTokens(
+      messages.slice(split).map((m) => getTextContent(m)).join("\n")
+    );
+    if (preservedTokens <= maxPreservedTokens) return split;
+  }
+  return findSafeSplitPoint(messages, keepRecentMessagesFallback);
+};
 var isCompactionSummary = (msg) => msg.metadata?.isCompactionSummary === true;
+var FAILURE_SIGNAL_RE = /\berror\b|\bfailed\b|\bfailure\b|\bexception\b|could ?n[’']?t|cannot\b|unable to|\[subagent result\]/i;
 var buildSummarizationMessages = (messagesToCompact, instructions) => {
   const hasPriorSummary = messagesToCompact.length > 0 && isCompactionSummary(messagesToCompact[0]);
-  const conversationLines = [];
+  const lines = [];
   for (let i = 0; i < messagesToCompact.length; i++) {
     const msg = messagesToCompact[i];
     const text = getTextContent(msg);
     const isPrior = i === 0 && hasPriorSummary;
     const rendered = isPrior || text.length <= SUMMARIZATION_MESSAGE_TRUNCATION_CHARS ? text : text.slice(0, SUMMARIZATION_MESSAGE_TRUNCATION_CHARS) + "\n...[truncated]";
     const tag = isPrior ? "prior-summary" : msg.role;
-    conversationLines.push(`[${tag}]: ${rendered}`);
+    const line = `[${tag}]: ${rendered}`;
+    lines.push({
+      text: line,
+      tokens: estimateTokens(line),
+      important: isPrior || FAILURE_SIGNAL_RE.test(text)
+    });
   }
+  let total = lines.reduce((sum, l) => sum + l.tokens, 0);
+  if (total > SUMMARIZATION_MAX_INPUT_TOKENS) {
+    for (let i = 0; i < lines.length && total > SUMMARIZATION_MAX_INPUT_TOKENS; ) {
+      if (lines[i].important) {
+        i++;
+        continue;
+      }
+      total -= lines[i].tokens;
+      lines.splice(i, 1);
+    }
+  }
+  const conversationLines = lines.map((l) => l.text);
   let prompt = SUMMARIZATION_PROMPT;
   if (hasPriorSummary) prompt = `${prompt}
 
@@ -475,7 +523,7 @@ var parseSubagentCallback = (msg) => {
   const subagentId = typeof meta.subagentId === "string" && meta.subagentId ? meta.subagentId : headerMatch?.[2] ?? "";
   if (!subagentId) return null;
   const task = typeof meta.task === "string" && meta.task ? meta.task : headerMatch?.[1] ?? "";
-  const status = headerMatch?.[3] ?? "completed";
+  const status = typeof meta.taskOutcome === "string" && meta.taskOutcome || headerMatch?.[3] || "unknown";
   const bodyStart = text.indexOf("\n\n");
   const body = bodyStart >= 0 ? text.slice(bodyStart + 2) : text;
   const digest = body.length > SUBAGENT_DIGEST_CHARS ? body.slice(0, SUBAGENT_DIGEST_CHARS) + "\u2026" : body;
@@ -535,8 +583,18 @@ ${summary}
 Continue from where the conversation left off without re-asking questions.`,
   metadata: { isCompactionSummary: true, timestamp: Date.now() }
 });
+var MAX_PRESERVED_CONTEXT_FRACTION = 0.5;
 var compactMessages = async (model, messages, config, options) => {
-  const splitIdx = findSafeSplitPoint(messages, config.keepRecentMessages);
+  const contextWindow = options?.contextWindow && options.contextWindow > 0 ? options.contextWindow : 2e5;
+  const maxPreservedTokens = Math.floor(
+    contextWindow * MAX_PRESERVED_CONTEXT_FRACTION
+  );
+  const splitIdx = findSafeSplitPointByTurns(
+    messages,
+    config.keepRecentTurns,
+    config.keepRecentMessages,
+    maxPreservedTokens
+  );
   if (splitIdx === -1) {
     return {
       compacted: false,
@@ -10790,7 +10848,11 @@ var AgentHarness = class _AgentHarness {
     const modelName = agent.frontmatter.model?.name ?? "claude-opus-4-5";
     const modelInstance = this.modelProvider(modelName);
     const config = resolveCompactionConfig(agent.frontmatter.compaction);
-    return compactMessages(modelInstance, messages, config, options);
+    const contextWindow = agent.frontmatter.model?.contextWindow ?? getModelContextWindow(modelName);
+    return compactMessages(modelInstance, messages, config, {
+      ...options,
+      contextWindow: options?.contextWindow ?? contextWindow
+    });
   }
   async *run(input) {
     if (!this.parsedAgent) {
@@ -10811,7 +10873,7 @@ var AgentHarness = class _AgentHarness {
     let agent = this.parsedAgent;
     const runId = `run_${randomUUID5()}`;
     const start = now();
-    const maxSteps = agent.frontmatter.limits?.maxSteps ?? 20;
+    const maxSteps = input.maxSteps ?? agent.frontmatter.limits?.maxSteps ?? 20;
     const configuredTimeout = this.runTimeoutSecOverride ?? agent.frontmatter.limits?.timeout;
     const timeoutMs = this.environment === "development" && configuredTimeout == null ? 0 : (configuredTimeout ?? 300) * 1e3;
     const platformMaxDurationSec = Number(process.env.PONCHO_MAX_DURATION) || 0;
@@ -11383,7 +11445,8 @@ ${textContent}` };
               const compactResult = await compactMessages(
                 modelInstance,
                 messages,
-                compactionConfig
+                compactionConfig,
+                { contextWindow }
               );
               if (compactResult.compacted) {
                 messages.length = 0;
@@ -12709,6 +12772,7 @@ var resolveRunRequest = (conversation, request) => {
 };
 
 // src/orchestrator/turn.ts
+import { randomUUID as randomUUID6 } from "crypto";
 var createTurnDraftState = () => ({
   assistantResponse: "",
   toolTimeline: [],
@@ -12930,6 +12994,52 @@ var buildApprovalCheckpoints = ({
   pendingToolCalls,
   kind
 }));
+var assembleCheckpointMessages = (conversation, checkpoint) => {
+  const n = normalizeApprovalCheckpoint(checkpoint, conversation.messages);
+  const base = n.baseMessageCount != null ? conversation.messages.slice(0, n.baseMessageCount) : [];
+  return [...base, ...n.checkpointMessages ?? []];
+};
+var buildToolResultMessage = (assistantMsg, toolResults) => {
+  if (assistantMsg?.role !== "assistant") return void 0;
+  let toolCalls = [];
+  try {
+    const parsed = JSON.parse(typeof assistantMsg.content === "string" ? assistantMsg.content : "");
+    toolCalls = parsed.tool_calls ?? [];
+  } catch {
+    return void 0;
+  }
+  if (toolCalls.length === 0) return void 0;
+  const provided = new Map(toolResults.map((r) => [r.callId, r]));
+  return {
+    role: "tool",
+    content: JSON.stringify(
+      toolCalls.map((tc) => {
+        const r = provided.get(tc.id);
+        return {
+          type: "tool_result",
+          tool_use_id: tc.id,
+          tool_name: r?.toolName ?? tc.name,
+          content: r ? r.error ? `Tool error: ${r.error}` : JSON.stringify(r.result ?? null) : "Tool error: Tool execution deferred (pending approval checkpoint)"
+        };
+      })
+    ),
+    metadata: { timestamp: Date.now(), id: randomUUID6() }
+  };
+};
+var buildResumeCheckpoints = ({
+  priorMessages,
+  checkpointEvent,
+  runId,
+  kind = "approval"
+}) => buildApprovalCheckpoints({
+  approvals: checkpointEvent.approvals,
+  runId,
+  checkpointMessages: [...priorMessages, ...checkpointEvent.checkpointMessages],
+  baseMessageCount: 0,
+  pendingToolCalls: checkpointEvent.pendingToolCalls,
+  kind
+});
+var messageText = (m) => typeof m.content === "string" ? m.content : Array.isArray(m.content) ? m.content.map((p) => p.text ?? "").join("") : "";
 var applyTurnMetadata = (conv, meta, opts = {}) => {
   const {
     clearContinuation = true,
@@ -12948,6 +13058,20 @@ var applyTurnMetadata = (conv, meta, opts = {}) => {
   } else if (shouldRebuildCanonical) {
     conv._harnessMessages = conv.messages;
   }
+  if (isMessageArray(conv._harnessMessages) && conv._harnessMessages.length > 0) {
+    const canonical = conv._harnessMessages;
+    const summarized = canonical.some((m) => m.metadata?.isCompactionSummary);
+    const lastUser = [...conv.messages].reverse().find((m) => m.role === "user");
+    const lastUserText = lastUser ? messageText(lastUser).trim() : "";
+    if (!summarized && lastUserText) {
+      const present = canonical.some((m) => m.role === "user" && messageText(m).trim() === lastUserText);
+      if (!present) {
+        console.error(
+          `[transcript-guard] conversation ${conv.conversationId}: model-facing transcript is missing the latest user message \u2014 it diverged from the display transcript. This is a resume/finalize message-assembly bug; the model will not see that turn's input.`
+        );
+      }
+    }
+  }
   if (meta.toolResultArchive !== void 0) {
     conv._toolResultArchive = meta.toolResultArchive;
   }
@@ -12978,12 +13102,12 @@ var STALE_SUBAGENT_THRESHOLD_MS = 5 * 60 * 1e3;
 import { createLogger as createLogger8, getTextContent as getTextContent3 } from "@poncho-ai/sdk";
 
 // src/orchestrator/entries-dual-write.ts
-import { randomUUID as randomUUID6 } from "crypto";
+import { randomUUID as randomUUID7 } from "crypto";
 var appendEntriesSafe = async (store, conversation, entries, log2) => {
   if (entries.length === 0) return [];
   try {
     const withIds = entries.map(
-      (e) => ({ id: randomUUID6(), ...e })
+      (e) => ({ id: randomUUID7(), ...e })
     );
     return await store.appendEntries(
       conversation.conversationId,
@@ -13037,6 +13161,17 @@ var abnormalEndResponse = (opts) => {
 
 ${opts.gathered}` : `${head} ${recover}`;
 };
+var SUBAGENT_OUTCOME_RE = /\[\[\s*OUTCOME\s*:\s*(succeeded|failed|partial)\s*\]\]/i;
+var SUBAGENT_OUTCOME_INSTRUCTION = "\n\nAt the very end of your final message, on its own line, append a machine-readable outcome verdict for the task you were given: `[[OUTCOME: succeeded]]`, `[[OUTCOME: partial]]`, or `[[OUTCOME: failed]]`, followed by a one-line reason. Judge honestly by whether YOU actually accomplished the task (produced the real result / wrote the files), not by whether the run finished. If a tool or capability you needed was missing, that is `failed`.";
+var deriveTaskOutcome = (gathered, abnormal) => {
+  if (abnormal) return "failed";
+  if (!gathered.trim()) return "failed";
+  const m = gathered.match(SUBAGENT_OUTCOME_RE);
+  if (!m) return "unknown";
+  const v = m[1].toLowerCase();
+  return v === "succeeded" ? "succeeded" : v === "partial" ? "partial" : "failed";
+};
+var stripOutcomeVerdict = (text) => text.replace(/\n*\[\[\s*OUTCOME\s*:\s*(?:succeeded|failed|partial)\s*\]\].*$/is, "").trimEnd();
 var AgentOrchestrator = class {
   harness;
   conversationStore;
@@ -13211,34 +13346,11 @@ var AgentOrchestrator = class {
     });
     let latestRunId = conversation.runtimeRunId ?? "";
     let checkpointedRun = false;
-    const normalizedCheckpoint = normalizeApprovalCheckpoint(checkpoint, conversation.messages);
-    const baseMessages = normalizedCheckpoint.baseMessageCount != null ? conversation.messages.slice(0, normalizedCheckpoint.baseMessageCount) : [];
-    const fullCheckpointMessages = [...baseMessages, ...normalizedCheckpoint.checkpointMessages];
-    let resumeToolResultMsg;
-    const lastCpMsg = fullCheckpointMessages[fullCheckpointMessages.length - 1];
-    if (lastCpMsg?.role === "assistant") {
-      try {
-        const parsed = JSON.parse(typeof lastCpMsg.content === "string" ? lastCpMsg.content : "");
-        const cpToolCalls = parsed.tool_calls ?? [];
-        if (cpToolCalls.length > 0) {
-          const providedMap = new Map(toolResults.map((r) => [r.callId, r]));
-          resumeToolResultMsg = {
-            role: "tool",
-            content: JSON.stringify(cpToolCalls.map((tc) => {
-              const provided = providedMap.get(tc.id);
-              return {
-                type: "tool_result",
-                tool_use_id: tc.id,
-                tool_name: provided?.toolName ?? tc.name,
-                content: provided ? provided.error ? `Tool error: ${provided.error}` : JSON.stringify(provided.result ?? null) : "Tool error: Tool execution deferred (pending approval checkpoint)"
-              };
-            })),
-            metadata: { timestamp: Date.now() }
-          };
-        }
-      } catch {
-      }
-    }
+    const fullCheckpointMessages = assembleCheckpointMessages(conversation, checkpoint);
+    const resumeToolResultMsg = buildToolResultMessage(
+      fullCheckpointMessages[fullCheckpointMessages.length - 1],
+      toolResults
+    );
     const fullCheckpointWithResults = resumeToolResultMsg ? [...fullCheckpointMessages, resumeToolResultMsg] : fullCheckpointMessages;
     let draftRef;
     let execution;
@@ -13283,12 +13395,10 @@ var AgentOrchestrator = class {
             const cpEvent = event;
             const conv = await this.conversationStore.get(conversationId);
             if (conv) {
-              conv.pendingApprovals = buildApprovalCheckpoints({
-                approvals: cpEvent.approvals,
-                runId: latestRunId,
-                checkpointMessages: [...fullCheckpointWithResults, ...cpEvent.checkpointMessages],
-                baseMessageCount: 0,
-                pendingToolCalls: cpEvent.pendingToolCalls
+              conv.pendingApprovals = buildResumeCheckpoints({
+                priorMessages: fullCheckpointWithResults,
+                checkpointEvent: cpEvent,
+                runId: latestRunId
               });
               conv.updatedAt = Date.now();
               await this.conversationStore.update(conv);
@@ -13480,6 +13590,7 @@ var AgentOrchestrator = class {
       subagentId,
       task: conv.subagentMeta?.task ?? conv.title,
       status: "completed",
+      taskOutcome: deriveTaskOutcome(responseText, false),
       result: { status: "completed", response: responseText, steps: 0, tokens: { input: 0, output: 0, cached: 0 }, duration: 0 },
       timestamp: Date.now()
     };
@@ -13581,7 +13692,11 @@ var AgentOrchestrator = class {
       const harnessMessages = [...runOutcome.messages];
       const recallParams = this.hooks?.buildRecallParams?.({ ownerId, tenantId: conversation.tenantId, excludeConversationId: childConversationId }) ?? {};
       for await (const event of childHarness.runWithTelemetry({
-        task,
+        // Ask the subagent to self-report a machine-readable task outcome at the
+        // end of its final message; `deriveTaskOutcome` parses it so a failed
+        // task can never be recorded as "completed". Appended only to the model
+        // input, not the stored/displayed task name.
+        task: `${task}${SUBAGENT_OUTCOME_INSTRUCTION}`,
         conversationId: childConversationId,
         tenantId: conversation.tenantId ?? void 0,
         parameters: withToolResultArchiveParam({
@@ -13768,11 +13883,14 @@ var AgentOrchestrator = class {
         if (freshSubConv) gathered = realResponseText(lastAssistantText(freshSubConv.messages));
       }
       const abnormal = !runResult;
-      const subagentResponse = abnormal ? abnormalEndResponse({ subagentId: childConversationId, gathered, runError }) : gathered;
+      const taskOutcome = deriveTaskOutcome(gathered, abnormal);
+      const cleanedGathered = stripOutcomeVerdict(gathered);
+      const subagentResponse = abnormal ? abnormalEndResponse({ subagentId: childConversationId, gathered: cleanedGathered, runError }) : cleanedGathered;
       const pendingResult = {
         subagentId: childConversationId,
         task,
         status: abnormal ? "error" : "completed",
+        taskOutcome,
         result: {
           status: runResult?.status ?? "error",
           response: subagentResponse,
@@ -13813,6 +13931,7 @@ var AgentOrchestrator = class {
         subagentId: childConversationId,
         task,
         status: "error",
+        taskOutcome: "failed",
         error: { code: "SUBAGENT_ERROR", message: errMsg },
         timestamp: Date.now()
       };
@@ -13880,10 +13999,10 @@ Response: ${responseLine}
 Steps: ${pr.result.steps}, Duration: ${pr.result.duration}ms` : pr.error ? `Error: ${pr.error.message}` : "(no result)";
       const injected = {
         role: "user",
-        content: `[Subagent Result] Subagent "${pr.task}" (${pr.subagentId}) ${pr.status}:
+        content: `[Subagent Result] Subagent "${pr.task}" (${pr.subagentId}) ${pr.taskOutcome}:
 
 ${resultBody}`,
-        metadata: { _subagentCallback: true, subagentId: pr.subagentId, task: pr.task, timestamp: pr.timestamp }
+        metadata: { _subagentCallback: true, subagentId: pr.subagentId, task: pr.task, taskOutcome: pr.taskOutcome, timestamp: pr.timestamp }
       };
       injectedCallbackMessages.push(injected);
       conversation.messages.push(injected);
@@ -14166,13 +14285,16 @@ ${resultBody}`,
         if (freshSubConv) gathered = realResponseText(lastAssistantText(freshSubConv.messages));
       }
       const abnormal = !runResult;
-      const subagentResponse = abnormal ? abnormalEndResponse({ subagentId: conversationId, gathered, runError }) : gathered;
+      const taskOutcome = deriveTaskOutcome(gathered, abnormal);
+      const cleanedGathered = stripOutcomeVerdict(gathered);
+      const subagentResponse = abnormal ? abnormalEndResponse({ subagentId: conversationId, gathered: cleanedGathered, runError }) : cleanedGathered;
       const parentConv = await this.conversationStore.get(parentConversationId);
       if (parentConv) {
         const result = {
           subagentId: conversationId,
           task,
           status: abnormal ? "error" : "completed",
+          taskOutcome,
           result: { status: runResult?.status ?? "error", response: subagentResponse, steps: runResult?.steps ?? 0, tokens: { input: 0, output: 0, cached: 0 }, duration: runResult?.duration ?? 0 },
           ...abnormal ? { error: { code: runError?.code ?? "SUBAGENT_INCOMPLETE", message: runError?.message ?? "subagent ended without a result" } } : {},
           timestamp: Date.now()
@@ -14221,6 +14343,7 @@ ${resultBody}`,
           subagentId: conversationId,
           task,
           status: "error",
+          taskOutcome: "failed",
           error: { code: "CONTINUATION_ERROR", message: err instanceof Error ? err.message : String(err) },
           timestamp: Date.now()
         };
@@ -14464,6 +14587,7 @@ ${resultBody}`,
       subagentId: childConversationId,
       task,
       status: "error",
+      taskOutcome: "failed",
       error: { code: "SUBAGENT_SPAWN_FAILED", message },
       timestamp: Date.now()
     });
@@ -14495,6 +14619,7 @@ ${resultBody}`,
             subagentId: conv.conversationId,
             task: conv.subagentMeta.task,
             status: "error",
+            taskOutcome: "failed",
             error: conv.subagentMeta.error,
             timestamp: Date.now()
           };
@@ -14521,7 +14646,7 @@ ${resultBody}`,
 };
 
 // src/orchestrator/run-conversation-turn.ts
-import { randomUUID as randomUUID7 } from "crypto";
+import { randomUUID as randomUUID8 } from "crypto";
 import { createLogger as createLogger9 } from "@poncho-ai/sdk";
 var log = createLogger9("orchestrator");
 var runConversationTurn = async (opts) => {
@@ -14561,9 +14686,9 @@ var runConversationTurn = async (opts) => {
   const userMessage = {
     role: "user",
     content: userContent,
-    metadata: { id: randomUUID7(), timestamp: turnTimestamp }
+    metadata: { id: randomUUID8(), timestamp: turnTimestamp }
   };
-  const assistantId = randomUUID7();
+  const assistantId = randomUUID8();
   const draft = createTurnDraftState();
   let latestRunId = conversation.runtimeRunId ?? "";
   let runCancelled = false;
@@ -14637,6 +14762,7 @@ var runConversationTurn = async (opts) => {
         disablePromptCache: opts.disablePromptCache,
         suppressTelemetry: opts.suppressTelemetry,
         model: opts.model,
+        maxSteps: opts.maxSteps,
         volatileContext: opts.volatileContext,
         telemetryAttributes: opts.telemetryAttributes
       },
@@ -14938,11 +15064,14 @@ export {
   abnormalEndResponse,
   appendEntriesSafe,
   applyTurnMetadata,
+  assembleCheckpointMessages,
   buildAgentDirectoryName,
   buildApprovalCheckpoints,
   buildAssistantMetadata,
+  buildResumeCheckpoints,
   buildSkillContextWindow,
   buildToolCompletedText,
+  buildToolResultMessage,
   cloneSections,
   compactMessages,
   completeOpenAICodexDeviceAuth,
@@ -14981,6 +15110,7 @@ export {
   estimateTotalTokens,
   executeConversationTurn,
   findSafeSplitPoint,
+  findSafeSplitPointByTurns,
   flushTurnDraft,
   generateAgentId,
   getAgentStoreDirectory,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/harness",
-  "version": "0.60.1",
+  "version": "0.61.1",
   "description": "Agent execution runtime - conversation loop, tool dispatch, streaming",
   "repository": {
     "type": "git",
@@ -34,7 +34,7 @@
     "mustache": "^4.2.0",
     "yaml": "^2.4.0",
     "zod": "^3.22.0",
-    "@poncho-ai/sdk": "1.16.0"
+    "@poncho-ai/sdk": "1.17.0"
   },
   "peerDependencies": {
     "esbuild": ">=0.17.0",

package/src/agent-parser.ts

@@ -52,6 +52,16 @@ export interface AgentFrontmatter {
 export 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;
 }
@@ -306,6 +316,10 @@ export const parseAgentMarkdown = (content: string): ParsedAgent => {
           "Invalid AGENT.md frontmatter compaction.trigger: must be between 0.1 and 1.",
         );
       }
+      const keepRecentTurns =
+        typeof raw.keepRecentTurns === "number"
+          ? Math.max(1, Math.floor(raw.keepRecentTurns))
+          : 4;
       const keepRecentMessages =
         typeof raw.keepRecentMessages === "number"
           ? Math.max(2, Math.floor(raw.keepRecentMessages))
@@ -314,7 +328,7 @@ export const parseAgentMarkdown = (content: string): ParsedAgent => {
         typeof raw.instructions === "string" && raw.instructions.trim()
           ? raw.instructions.trim()
           : undefined;
-      return { enabled, trigger, keepRecentMessages, instructions };
+      return { enabled, trigger, keepRecentTurns, keepRecentMessages, instructions };
     })(),
     cron: parseCronJobs(parsed.cron),
   };