@sema-agent/core 2.13.0 → 4.0.0

package/dist/core/runner/runtask.js

@@ -9,8 +9,7 @@ import { ASK_USER_QUESTION_TOOL_NAME } from "../ask-question.js";
 import { computeCostMicroUsd, modelCostToPricing } from "../pricing.js";
 import { emitTrace } from "../trace.js";
 import { emitTaskOutcome } from "../task-outcome.js";
-import { DEGENERATE_MESSAGE } from "../../brain/repetition.js";
-import { WALLTIME_CUTOFF_MESSAGE } from "../../brain/walltime.js";
+import { isDegenerateCutMessage, isWalltimeCutMessage } from "../../brain/terminal-cause.js";
 import { primaryActivityArg } from "../arg-summary.js";
 import { resolveReasoning } from "../../brain/reasoning.js";
 import { readDegradation } from "../../brain/degrading.js";
@@ -24,8 +23,10 @@ import { toImageContent } from "./image.js";
 import { OUTPUT_TOOL_NAME, SKILLS_LISTING_PROBE_HEADER, resolveOutputRetries } from "./synthetic-tools.js";
 import { cacheFamilyOf, usageCostMicroUsd } from "./usage-accounting.js";
 import { assembleResult, errorCodeOf } from "./assemble-result.js";
-import { ATTACHMENT_BYTE_CAP, CHANGED_FILES_MAX, AGENT_LISTING_REMOVED_HEADER, SKILLS_LISTING_DELTA_HEADER, SKILLS_LISTING_REMOVED_HEADER, advanceCadenceClock, agentListingDeltaHeader, agentListingInitialHeader, replayAnnouncedListing, replayAnnouncedModels, collectDateChange, collectDueAttachments, collectInstructionsChange, commitAgentListing, commitInstructionsChange, commitSkillsListing, createAttachmentState, rebaseCadenceWindows, reduceToolEnd, renderAgentListingDelta, renderMcpDroppedTools, renderMcpInstructionsDelta, renderOrphanedBackgroundTasks, selectMcpDroppedBatch, renderSkillsListingDelta, renderToolsDelta, stampWriteAnchor } from "./turn-attachments.js";
+import { ATTACHMENT_BYTE_CAP, CHANGED_FILES_MAX, AGENT_LISTING_REMOVED_HEADER, SKILLS_LISTING_DELTA_HEADER, SKILLS_LISTING_REMOVED_HEADER, advanceCadenceClock, agentListingDeltaHeader, agentListingInitialHeader, replayAnnouncedListing, replayAnnouncedModels, clipToBytes, collectDateChange, collectDueAttachments, collectInstructionsChange, commitAgentListing, commitInstructionsChange, commitSkillsListing, createAttachmentState, rebaseCadenceWindows, reduceToolEnd, renderAgentListingDelta, renderMcpDroppedTools, renderMcpInstructionsDelta, renderOrphanedBackgroundTasks, selectMcpDroppedBatch, renderSkillsListingDelta, renderToolsDelta, stampWriteAnchor } from "./turn-attachments.js";
+import { buildWorkingFileAttachments, centerAdoptionOption, emitInputTruncated } from "./compaction-call-options.js";
 import { prepareTask } from "./prepare-task.js";
+import { settleTeardownLeg } from "./teardown-bounded.js";
 import { TOOL_SEARCH_NAME } from "./tool-disclosure.js";
 import { hasVerifiableStructureSignal } from "./grounding-signal.js";
 import { hasDestroy, isIsolated } from "../remote-env.js";
@@ -87,20 +88,6 @@ function batchResponseDigest(result) {
         ? `${text.slice(0, BATCH_RESPONSE_MAX_CHARS)}…[+${text.length - BATCH_RESPONSE_MAX_CHARS} chars truncated]`
         : text;
 }
-function clipToByteBudget(s, maxBytes) {
-    if (Buffer.byteLength(s, "utf8") <= maxBytes)
-        return s;
-    let lo = 0;
-    let hi = s.length;
-    while (lo < hi) {
-        const mid = (lo + hi + 1) >> 1;
-        if (Buffer.byteLength(s.slice(0, mid), "utf8") <= maxBytes)
-            lo = mid;
-        else
-            hi = mid - 1;
-    }
-    return s.slice(0, lo);
-}
 const BATCH_CONTEXT_MIN_KEEP_BYTES = 160;
 const BATCH_TRUNCATION_MARKER = "\n…[truncated]";
 const SUGGESTIONS_DEFAULT_COUNT = 3;
@@ -118,6 +105,10 @@ function resolveMaxTurns(limits) {
     }
     return DEFAULT_MAX_TURNS;
 }
+function resumeDecisionWasNegative(resume) {
+    const o = resume.outcome;
+    return o.decision === "reject" || o.decision === "deny";
+}
 const DEFERRED_REISSUE = "[DEFERRED] This tool call shared a batch with a call that suspended for durable approval, so it was " +
     "NOT executed on resume. If you still need it, issue it again now.";
 function deepJsonEqual(a, b) {
@@ -341,8 +332,7 @@ function makeTurnBoundary(prepared, stats, rs, deps) {
                     const scan = await Promise.race([
                         prepared.detectExternalChanges(maxFiles),
                         new Promise((resolve) => {
-                            const t = setTimeout(() => resolve({ changed: [], evicted: [] }), SCAN_CEILING_MS);
-                            t.unref?.();
+                            setTimeout(() => resolve({ changed: [], evicted: [] }), SCAN_CEILING_MS);
                         }),
                     ]);
                     for (const p of scan.evicted)
@@ -475,8 +465,7 @@ function makeTurnBoundary(prepared, stats, rs, deps) {
                     probed = await Promise.race([
                         rs.attach.instrProbe(prepared.instructionSources),
                         new Promise((resolve) => {
-                            const t = setTimeout(() => resolve(null), PROBE_CEILING_MS);
-                            t.unref?.();
+                            setTimeout(() => resolve(null), PROBE_CEILING_MS);
                         }),
                     ]);
                 }
@@ -506,16 +495,17 @@ function makeTurnBoundary(prepared, stats, rs, deps) {
             const batch = rs.turn.toolBatch;
             rs.turn.toolBatch = [];
             batchArgs?.clear();
-            if (!boundarySteered && !rs.counters.finalizeInjected && !finalVerifyInjectedThisTurn && !prepared.abortController.signal.aborted) {
+            const injectedThisTurn = finalVerifyInjectedThisTurn ? "final_verification" : rs.counters.finalizeInjected ? "finalize" : undefined;
+            if (!boundarySteered && !prepared.abortController.signal.aborted) {
                 try {
-                    const r = await postToolBatchHook(batch);
-                    if (r?.additionalContext) {
+                    const r = await postToolBatchHook(batch, injectedThisTurn !== undefined ? { injectedThisTurn } : undefined);
+                    if (r?.additionalContext && injectedThisTurn === undefined) {
                         const body = sanitizeUntrustedText(r.additionalContext);
                         const budget = ATTACHMENT_BYTE_CAP - boundaryAttachmentBytes;
                         if (budget >= BATCH_CONTEXT_MIN_KEEP_BYTES) {
-                            const clipped = clipToByteBudget(body, budget) === body
+                            const clipped = clipToBytes(body, budget) === body
                                 ? body
-                                : `${clipToByteBudget(body, budget - Buffer.byteLength(BATCH_TRUNCATION_MARKER, "utf8"))}${BATCH_TRUNCATION_MARKER}`;
+                                : `${clipToBytes(body, budget - Buffer.byteLength(BATCH_TRUNCATION_MARKER, "utf8"))}${BATCH_TRUNCATION_MARKER}`;
                             batchContextBlock = `<system-reminder>\n${clipped}\n</system-reminder>`;
                         }
                     }
@@ -605,10 +595,7 @@ function makeTurnBoundary(prepared, stats, rs, deps) {
             const comp = await maybeCompact({
                 session: prepared.session,
                 epochDeclaredSections: prepared.epochDeclaredSections,
-                ...(() => {
-                    const ca = prepared.centerCompactionCandidate?.();
-                    return ca !== undefined ? { centerAdoption: ca } : {};
-                })(),
+                ...centerAdoptionOption(prepared),
                 model: event.model,
                 compactionModel: prepared.compModel,
                 brain: compactionBrain,
@@ -620,22 +607,11 @@ function makeTurnBoundary(prepared, stats, rs, deps) {
                 ...(manualCancelSignal !== undefined ? { manualCancelSignal } : {}),
                 minTokens: forceManual ? 0 : rs.counters.compactionFloor,
                 force: forceManual,
-                onInputTruncated: (info) => emitTrace(rs.telemetry.tracer, () => ({ kind: "compaction.input_truncated", version: 1, taskId: rs.telemetry.taskId, label: info.label, droppedChars: info.droppedChars, keptChars: info.keptChars, ts: Date.now() })),
+                onInputTruncated: emitInputTruncated(rs.telemetry.tracer, rs.telemetry.taskId),
                 forceUnderThreshold: trimPressure,
                 overheadTokens: prepared.promptOverheadTokens,
                 ...(prepared.activeTools.size > 0 ? { activeTools: [...prepared.activeTools] } : {}),
-                workingFileAttachments: spec.compaction?.attachWorkingFiles !== false && prepared.readTaskFile
-                    ? {
-                        readFile: prepared.readTaskFile,
-                        ...(prepared.recentlyReadFiles ? { recentlyReadFiles: prepared.recentlyReadFiles } : {}),
-                        ...(prepared.normalizeAttachmentPath ? { normalizePath: prepared.normalizeAttachmentPath } : {}),
-                        ...(prepared.isDedupStubResult ? { isDedupStubResult: prepared.isDedupStubResult } : {}),
-                        ...(prepared.instructionSources
-                            ? { instructionSourcePaths: prepared.instructionSources.filter((s) => s.contentHash !== null).map((s) => s.path) }
-                            : {}),
-                        ...(typeof spec.compaction?.attachWorkingFiles === "object" ? spec.compaction.attachWorkingFiles : undefined),
-                    }
-                    : undefined,
+                workingFileAttachments: buildWorkingFileAttachments(spec, prepared),
                 ...(prepared.onCompactionApplied ? { onApplied: prepared.onCompactionApplied } : {}),
                 ...runnerHooks.seamCCompactionOptions(prepared),
                 ...windowSafetyOptions(event.model),
@@ -811,16 +787,18 @@ function makeHarnessHandlers(prepared, stats, rs, deps) {
                 rs.turn.turnUsageMissing = true;
             const u = m.usage;
             stats.tokens += u.totalTokens || 0;
-            const { totalInputTokens: turnInput, costMicroUsd: turnCostMicroUsd } = usageCostMicroUsd(rs.telemetry.cacheFamily, u, rs.telemetry.pricing);
+            const { totalInputTokens: turnTotal, uncachedInputTokens: turnUncached, costMicroUsd: turnCostMicroUsd, } = usageCostMicroUsd(rs.telemetry.cacheFamily, u, rs.telemetry.pricing);
             if (!rs.telemetry.pricingConfigured)
                 rs.telemetry.unpricedSpend = true;
-            stats.promptTokens += turnInput;
+            stats.promptTokens += turnUncached;
+            stats.totalInputTokens += turnTotal;
             stats.cachedTokens += u.cacheRead || 0;
             stats.cacheWriteTokens += u.cacheWrite || 0;
             stats.outputTokens += u.output || 0;
             stats.costMicroUsd += turnCostMicroUsd;
-            rs.turn.turnUsage ??= { inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheWriteTokens: 0, costMicroUsd: 0 };
-            rs.turn.turnUsage.inputTokens += turnInput;
+            rs.turn.turnUsage ??= { inputTokens: 0, totalInputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheWriteTokens: 0, costMicroUsd: 0 };
+            rs.turn.turnUsage.inputTokens += turnUncached;
+            rs.turn.turnUsage.totalInputTokens += turnTotal;
             rs.turn.turnUsage.outputTokens += u.output || 0;
             rs.turn.turnUsage.cacheReadTokens += u.cacheRead || 0;
             rs.turn.turnUsage.cacheWriteTokens += u.cacheWrite || 0;
@@ -846,7 +824,8 @@ function makeHarnessHandlers(prepared, stats, rs, deps) {
                 model: m.model,
                 provider: m.provider,
                 turn: stats.turns + 1,
-                promptTokens: turnInput,
+                promptTokens: turnUncached,
+                totalInputTokens: turnTotal,
                 completionTokens: u.output || 0,
                 cacheRead: u.cacheRead || 0,
                 cacheWrite: u.cacheWrite || 0,
@@ -890,7 +869,7 @@ function makeHarnessHandlers(prepared, stats, rs, deps) {
         if (m.role === "assistant" && typeof m.stopReason === "string") {
             rs.turn.turnStopReason = m.stopReason;
         }
-        if (m.role === "assistant" && m.errorMessage === WALLTIME_CUTOFF_MESSAGE) {
+        if (m.role === "assistant" && isWalltimeCutMessage(m)) {
             rs.counters.callCutoffs += 1;
             if (prepared.callCapRef)
                 prepared.callCapRef.finalizeMode = true;
@@ -1262,7 +1241,6 @@ export class Runner {
             let t;
             const timeout = new Promise((res2) => {
                 t = setTimeout(() => res2(undefined), READY_TIMEOUT_MS);
-                t.unref?.();
             });
             return Promise.race([p, timeout]).finally(() => clearTimeout(t));
         };
@@ -1313,7 +1291,7 @@ export class Runner {
         const settled = run.catch(async (err) => {
             let code = errorCodeOf(err);
             const reopenReason = code === "resume.tool_unavailable" ? "tool_unavailable" : "env_failed";
-            const reopenable = resume !== undefined && resume.pendingActionStarted !== true && reopenReason !== undefined;
+            const reopenable = resume !== undefined && resume.pendingActionStarted !== true && reopenReason !== undefined && !(resumeDecisionWasNegative(resume) && resume.decisionDelivered === true);
             let reopenCommitted = false;
             if (reopenable && resume?.onEnvRestoreFailed) {
                 try {
@@ -1333,11 +1311,7 @@ export class Runner {
             }
             if (resume && !reopenCommitted) {
                 this.parentConstraintRegistry.delete(resume.cp.token);
-                try {
-                    await this.sessions.unpin?.(resume.cp.sessionId);
-                }
-                catch {
-                }
+                await settleTeardownLeg(() => this.sessions.unpin?.(resume.cp.sessionId), "sessions.unpin (resume-failure leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: resume.cp.sessionId }));
             }
             if (resultValue !== undefined) {
                 try {
@@ -1650,7 +1624,7 @@ export class Runner {
             }
         }
         onReady({ harness: prepared.harness, abortController: prepared.abortController });
-        const stats = { turns: 0, tokens: 0, toolCalls: 0, promptTokens: 0, cachedTokens: 0, cacheWriteTokens: 0, cacheWriteTokensLong: 0, outputTokens: 0, costMicroUsd: 0 };
+        const stats = { turns: 0, tokens: 0, toolCalls: 0, promptTokens: 0, totalInputTokens: 0, cachedTokens: 0, cacheWriteTokens: 0, cacheWriteTokensLong: 0, outputTokens: 0, costMicroUsd: 0 };
         prepared.liveSpendRef.get = () => ({ costMicroUsd: stats.costMicroUsd, tokens: stats.tokens, turns: stats.turns, walltimeMs: Math.round(performance.now() - rs.telemetry.taskStartMonotonic) });
         if (resume &&
             resume.cp.suspendedAt !== undefined &&
@@ -2132,9 +2106,10 @@ export class Runner {
             const priced = this.deps.pricing?.[m.id] !== undefined || m.cost !== undefined;
             if (!priced)
                 rs.telemetry.unpricedSpend = true;
-            const { totalInputTokens, costMicroUsd } = usageCostMicroUsd(fam, u, price);
+            const { totalInputTokens, uncachedInputTokens, costMicroUsd } = usageCostMicroUsd(fam, u, price);
             stats.tokens += u.totalTokens || 0;
-            stats.promptTokens += totalInputTokens;
+            stats.promptTokens += uncachedInputTokens;
+            stats.totalInputTokens += totalInputTokens;
             stats.cachedTokens += u.cacheRead || 0;
             stats.cacheWriteTokens += u.cacheWrite || 0;
             stats.outputTokens += u.output || 0;
@@ -2142,7 +2117,7 @@ export class Runner {
             stats.compactionMicroUsd = (stats.compactionMicroUsd ?? 0) + costMicroUsd;
             emitTrace(rs.telemetry.tracer, () => ({
                 kind: "brain.call", version: 1, taskId: rs.telemetry.taskId, model: m.id, provider: m.provider,
-                promptTokens: totalInputTokens, completionTokens: u.output || 0,
+                promptTokens: uncachedInputTokens, totalInputTokens, completionTokens: u.output || 0,
                 cacheRead: u.cacheRead || 0, cacheWrite: u.cacheWrite || 0,
                 latencyMs: 0, ...(priced ? { costMicroUsd } : {}), ...(typeof msg?.stopReason === "string" ? { stopReason: msg.stopReason } : {}), ts: Date.now(),
             }));
@@ -2323,9 +2298,9 @@ export class Runner {
             truncatedOutput: {},
             malformedToolUse: {},
             thinkingOnly: {},
-            degenerateOutput: { detect: (m) => m.errorMessage === DEGENERATE_MESSAGE },
+            degenerateOutput: { detect: (m) => isDegenerateCutMessage(m) },
             walltimeCutoff: {
-                detect: (m) => m.errorMessage === WALLTIME_CUTOFF_MESSAGE,
+                detect: (m) => isWalltimeCutMessage(m),
                 hasLiveWriteoutWindow: () => {
                     const ref = prepared.callCapRef;
                     if (ref === undefined)
@@ -2348,10 +2323,7 @@ export class Runner {
                         const comp = await maybeCompact({
                             session: prepared.session,
                             epochDeclaredSections: prepared.epochDeclaredSections,
-                            ...(() => {
-                                const ca = prepared.centerCompactionCandidate?.();
-                                return ca !== undefined ? { centerAdoption: ca } : {};
-                            })(),
+                            ...centerAdoptionOption(prepared),
                             model: prepared.harness.getModel(),
                             compactionModel: prepared.compModel,
                             brain: compactionBrain,
@@ -2364,19 +2336,8 @@ export class Runner {
                             force: true,
                             overheadTokens: prepared.promptOverheadTokens,
                             ...(prepared.activeTools.size > 0 ? { activeTools: [...prepared.activeTools] } : {}),
-                            onInputTruncated: (info) => emitTrace(rs.telemetry.tracer, () => ({ kind: "compaction.input_truncated", version: 1, taskId: rs.telemetry.taskId, label: info.label, droppedChars: info.droppedChars, keptChars: info.keptChars, ts: Date.now() })),
-                            workingFileAttachments: spec.compaction?.attachWorkingFiles !== false && prepared.readTaskFile
-                                ? {
-                                    readFile: prepared.readTaskFile,
-                                    ...(prepared.recentlyReadFiles ? { recentlyReadFiles: prepared.recentlyReadFiles } : {}),
-                                    ...(prepared.normalizeAttachmentPath ? { normalizePath: prepared.normalizeAttachmentPath } : {}),
-                                    ...(prepared.isDedupStubResult ? { isDedupStubResult: prepared.isDedupStubResult } : {}),
-                                    ...(prepared.instructionSources
-                                        ? { instructionSourcePaths: prepared.instructionSources.filter((s) => s.contentHash !== null).map((s) => s.path) }
-                                        : {}),
-                                    ...(typeof spec.compaction?.attachWorkingFiles === "object" ? spec.compaction.attachWorkingFiles : undefined),
-                                }
-                                : undefined,
+                            onInputTruncated: emitInputTruncated(rs.telemetry.tracer, rs.telemetry.taskId),
+                            workingFileAttachments: buildWorkingFileAttachments(spec, prepared),
                             ...(prepared.onCompactionApplied ? { onApplied: prepared.onCompactionApplied } : {}),
                             ...this.seamCCompactionOptions(prepared),
                             ...windowSafetyOptions(prepared.harness.getModel()),
@@ -2498,6 +2459,8 @@ export class Runner {
                             resume.pendingActionStarted = true;
                         }
                     });
+                    if (resume.outcome.gate === "policy_ask")
+                        resume.decisionDelivered = true;
                 }
                 if (walltimeExhaustedResume) {
                     timeout.fired = true;
@@ -2529,6 +2492,8 @@ export class Runner {
                         if (steer.trusted)
                             engineSegments.push({ start, end: continuation.length });
                     }
+                    if (resume !== undefined)
+                        resume.decisionDelivered = true;
                     final = await withBrainSinks(() => prepared.harness.prompt(continuation, { engineSegments }));
                 }
             }
@@ -2639,17 +2604,8 @@ export class Runner {
         }
         catch (err) {
             if (errorCodeOf(err) === "resume.tool_unavailable") {
-                try {
-                    await prepared.mcp.dispose();
-                }
-                catch {
-                }
-                try {
-                    if (prepared.ownedEnv && hasDestroy(prepared.ownedEnv))
-                        await prepared.ownedEnv.destroy();
-                }
-                catch {
-                }
+                await settleTeardownLeg(() => prepared.mcp.dispose(), "mcp.dispose (tool_unavailable leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
+                await settleTeardownLeg(() => (prepared.ownedEnv && hasDestroy(prepared.ownedEnv) ? prepared.ownedEnv.destroy() : undefined), "ownedEnv.destroy (tool_unavailable leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
                 throw err;
             }
             threw = err;
@@ -2738,20 +2694,20 @@ export class Runner {
                     tokens: prepared.nestedStats.tokens,
                     turns: prepared.nestedStats.turns,
                     tasks: prepared.nestedStats.tasks,
-                    costMicroUsd: prepared.nestedStats.costMicroUsd || undefined,
+                    ...(prepared.nestedStats.anyUnpriced ? {} : { costMicroUsd: prepared.nestedStats.costMicroUsd }),
                 };
             }
-            if (stats.promptTokens > 0) {
-                const rawHit = stats.cachedTokens / stats.promptTokens;
-                if (stats.cachedTokens > stats.promptTokens) {
-                    this.deps.onError?.(new Error(`prompt-cache: usage/api mismatch — cachedTokens (${stats.cachedTokens}) > promptTokens (${stats.promptTokens}); the cache family is likely wrong. Does model.api match the brain serving it? Override with model.params.promptCacheFamily = "input-includes-cached" | "input-excludes-cached".`), { phase: "prompt-cache", sessionId: prepared.sessionId });
+            if (stats.totalInputTokens > 0) {
+                const rawHit = stats.cachedTokens / stats.totalInputTokens;
+                if (stats.cachedTokens > stats.totalInputTokens) {
+                    this.deps.onError?.(new Error(`prompt-cache: usage/api mismatch — cachedTokens (${stats.cachedTokens}) > totalInputTokens (${stats.totalInputTokens}); the cache family is likely wrong. Does model.api match the brain serving it? Override with model.params.promptCacheFamily = "input-includes-cached" | "input-excludes-cached".`), { phase: "prompt-cache", sessionId: prepared.sessionId });
                 }
                 stats.cacheHitRate = Math.min(1, rawHit);
-                if (!rs.telemetry.cacheBreakReported && stats.turns >= 2 && stats.promptTokens >= 8000 && stats.cacheHitRate < 0.15) {
+                if (!rs.telemetry.cacheBreakReported && stats.turns >= 2 && stats.totalInputTokens >= 8000 && stats.cacheHitRate < 0.15) {
                     const cause = rs.degrade.degraded
                         ? `a mid-task model switch (${rs.degrade.degraded.from} → ${rs.degrade.degraded.to}, degradation) reset the prefix cache — this is the likely cause`
                         : `the cacheable prefix looks unstable. Keep volatile content (memory/timestamps/ids) out of the prompt prefix and tool order stable, or it's a server-side/TTL miss`;
-                    this.deps.onError?.(new Error(`prompt-cache: low prefix-cache hit rate ${(stats.cacheHitRate * 100).toFixed(0)}% over ${stats.turns} turns (${stats.promptTokens} prompt tokens) — ${cause}. See design/09.`), { phase: "prompt-cache", sessionId: prepared.sessionId });
+                    this.deps.onError?.(new Error(`prompt-cache: low prefix-cache hit rate ${(stats.cacheHitRate * 100).toFixed(0)}% over ${stats.turns} turns (${stats.totalInputTokens} prompt tokens) — ${cause}. See design/09.`), { phase: "prompt-cache", sessionId: prepared.sessionId });
                 }
             }
             if (prepared.resourceLedger &&
@@ -2823,6 +2779,8 @@ export class Runner {
                 resume.pendingActionStarted !== true &&
                 result.status !== "completed" &&
                 result.status !== "suspended" &&
+                result.status !== "needs_review" &&
+                !(resumeDecisionWasNegative(resume) && resume.decisionDelivered === true) &&
                 resume.onEnvRestoreFailed !== undefined) {
                 try {
                     await resume.onEnvRestoreFailed("env_failed");
@@ -2843,15 +2801,11 @@ export class Runner {
                     const curWt = prepared.worktreeSessionRef?.current;
                     const changed = curCwd !== wss.restoredHandsCwd || curWt?.worktreeDir !== wss.restoredWorktreeDir;
                     if (curCwd !== undefined || curWt !== undefined || changed || wss.baselineUnknown === true) {
-                        try {
-                            await prepared.session.appendWorkspaceState({
-                                taskRoot: wss.rootCanonical,
-                                ...(curCwd !== undefined ? { handsCwd: curCwd } : {}),
-                                ...(curWt !== undefined ? { activeWorktree: { ...curWt } } : {}),
-                            });
-                        }
-                        catch {
-                        }
+                        await settleTeardownLeg(() => prepared.session.appendWorkspaceState({
+                            taskRoot: wss.rootCanonical,
+                            ...(curCwd !== undefined ? { handsCwd: curCwd } : {}),
+                            ...(curWt !== undefined ? { activeWorktree: { ...curWt } } : {}),
+                        }), "session.appendWorkspaceState (settle leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
                     }
                 }
             }
@@ -2859,8 +2813,8 @@ export class Runner {
             if (stopFailureHook !== undefined &&
                 result.status === "failed" &&
                 final?.stopReason === "error" &&
-                final.errorMessage !== DEGENERATE_MESSAGE &&
-                final.errorMessage !== WALLTIME_CUTOFF_MESSAGE &&
+                !isDegenerateCutMessage(final) &&
+                !isWalltimeCutMessage(final) &&
                 threw === undefined &&
                 rs.limits.budgetHit === undefined &&
                 !rs.degrade.outputInvalid &&
@@ -2901,19 +2855,12 @@ export class Runner {
                 }
             }
             if (resume && committedToken === undefined) {
-                try {
-                    await this.sessions.unpin?.(resume.cp.sessionId);
-                }
-                catch (err) {
-                    try {
-                        this.deps.onError?.(err, { phase: "config", sessionId: prepared.sessionId });
-                    }
-                    catch {
-                    }
-                }
+                await settleTeardownLeg(() => this.sessions.unpin?.(resume.cp.sessionId), "sessions.unpin (resumed-terminal leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
+            }
+            if (prepared.memoryEngineSession) {
+                const mes = prepared.memoryEngineSession;
+                await settleTeardownLeg(() => mes.harvest(), "memoryEngineSession.harvest (settle leg)", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
             }
-            if (prepared.memoryEngineSession)
-                await prepared.memoryEngineSession.harvest();
             emitTrace(rs.telemetry.tracer, () => ({
                 kind: "task.end",
                 version: 1,
@@ -2967,28 +2914,19 @@ export class Runner {
                 isIsolated(prepared.ownedEnv) &&
                 prepared.suspendRef.token === undefined &&
                 prepared.reviewRef.token === undefined;
-            try {
-                await defaultTaskRegistry.settleKilledForOwner({ owner: spec.taskId ?? prepared.sessionId, scope: spec.principal ?? "default", sessionId: prepared.sessionId }, {
-                    source: abortedLive &&
-                        !timeout.fired &&
-                        !rs.limits.turnsExceeded &&
-                        rs.limits.budgetHit === undefined &&
-                        prepared.suspendRef.token === undefined &&
-                        prepared.reviewRef.token === undefined
-                        ? "user"
-                        : "parent",
-                    skipSessionScoped: true,
-                    envDying: ownedEnvDying,
-                    retainProcesses: spec.retainBackgroundProcesses === true,
-                });
-            }
-            catch (e) {
-                try {
-                    this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId });
-                }
-                catch {
-                }
-            }
+            await settleTeardownLeg(() => defaultTaskRegistry.settleKilledForOwner({ owner: spec.taskId ?? prepared.sessionId, scope: spec.principal ?? "default", sessionId: prepared.sessionId }, {
+                source: abortedLive &&
+                    !timeout.fired &&
+                    !rs.limits.turnsExceeded &&
+                    rs.limits.budgetHit === undefined &&
+                    prepared.suspendRef.token === undefined &&
+                    prepared.reviewRef.token === undefined
+                    ? "user"
+                    : "parent",
+                skipSessionScoped: true,
+                envDying: ownedEnvDying,
+                retainProcesses: spec.retainBackgroundProcesses === true,
+            }), "taskRegistry.settleKilledForOwner", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
             const bgEnv = prepared.ownedEnv ?? this.deps.executionEnv;
             const retainBgBySpec = spec.retainBackgroundProcesses === true && !ownedEnvDying;
             if (bgEnv && hasBackgroundShell(bgEnv) && !retainBgBySpec) {
@@ -3000,7 +2938,7 @@ export class Runner {
                             ...defaultTaskRegistry.timeoutResidentShellIds(bgEnv),
                             ...defaultTaskRegistry.retainedShellIds(bgEnv),
                         ];
-                    await sweepBackgroundShells(bgEnv, defaultTaskRegistry, keepAlive.length > 0 ? { except: keepAlive } : undefined);
+                    await settleTeardownLeg(() => sweepBackgroundShells(bgEnv, defaultTaskRegistry, keepAlive.length > 0 ? { except: keepAlive } : undefined), "sweepBackgroundShells", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
                 }
                 catch (e) {
                     this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId });
@@ -3015,14 +2953,9 @@ export class Runner {
                 scope: spec.principal ?? "default",
             }, { skipSessionScoped: true });
             if (prepared.subagentRetain) {
-                try {
-                    await prepared.subagentRetain.disposeAll();
-                }
-                catch (e) {
-                    this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId });
-                }
+                await settleTeardownLeg(() => prepared.subagentRetain.disposeAll(), "subagentRetain.disposeAll", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
             }
-            await this.teardownOwnedEnv(prepared);
+            await settleTeardownLeg(() => this.teardownOwnedEnv(prepared), "teardownOwnedEnv", (e) => this.deps.onError?.(e, { phase: "config", sessionId: prepared.sessionId }));
         }
     }
     async suggestNextPrompts(spec, prepared, result) {
@@ -3035,7 +2968,6 @@ export class Runner {
                 ac.abort();
                 resolve([]);
             }, SUGGESTIONS_TIMEOUT_MS);
-            timer.unref?.();
         });
         const work = (async () => {
             let model = prepared.model;
@@ -3107,7 +3039,6 @@ export class Runner {
                     ac.abort();
                     resolve({ ok: false, error: { code: "aborted", message: "snapshot timed out" } });
                 }, SNAPSHOT_TIMEOUT_MS);
-                timer.unref?.();
             });
             const work = store.snapshot(prepared.sessionId, leafId, env, root, ac.signal);
             work.catch(() => { });
@@ -3526,10 +3457,7 @@ export class Runner {
                 const finishComp = await maybeCompact({
                     session: prepared.session,
                     epochDeclaredSections: prepared.epochDeclaredSections,
-                    ...(() => {
-                        const ca = prepared.centerCompactionCandidate?.();
-                        return ca !== undefined ? { centerAdoption: ca } : {};
-                    })(),
+                    ...centerAdoptionOption(prepared),
                     model: prepared.model,
                     compactionModel: prepared.compModel,
                     brain: opts?.brain ?? this.deps.brain,
@@ -3540,28 +3468,9 @@ export class Runner {
                     minTokens: opts?.minTokens,
                     overheadTokens: prepared.promptOverheadTokens,
                     ...(prepared.activeTools.size > 0 ? { activeTools: [...prepared.activeTools] } : {}),
-                    onInputTruncated: (info) => emitTrace(spec.tracer ?? this.deps.tracer, () => ({
-                        kind: "compaction.input_truncated",
-                        version: 1,
-                        taskId: spec.taskId ?? prepared.sessionId,
-                        label: info.label,
-                        droppedChars: info.droppedChars,
-                        keptChars: info.keptChars,
-                        ts: Date.now(),
-                    })),
+                    onInputTruncated: emitInputTruncated(spec.tracer ?? this.deps.tracer, spec.taskId ?? prepared.sessionId),
                     ...(opts?.windowSafety ?? {}),
-                    workingFileAttachments: spec.compaction?.attachWorkingFiles !== false && prepared.readTaskFile
-                        ? {
-                            readFile: prepared.readTaskFile,
-                            ...(prepared.recentlyReadFiles ? { recentlyReadFiles: prepared.recentlyReadFiles } : {}),
-                            ...(prepared.normalizeAttachmentPath ? { normalizePath: prepared.normalizeAttachmentPath } : {}),
-                            ...(prepared.isDedupStubResult ? { isDedupStubResult: prepared.isDedupStubResult } : {}),
-                            ...(prepared.instructionSources
-                                ? { instructionSourcePaths: prepared.instructionSources.filter((s) => s.contentHash !== null).map((s) => s.path) }
-                                : {}),
-                            ...(typeof spec.compaction?.attachWorkingFiles === "object" ? spec.compaction.attachWorkingFiles : undefined),
-                        }
-                        : undefined,
+                    workingFileAttachments: buildWorkingFileAttachments(spec, prepared),
                     ...(prepared.onCompactionApplied ? { onApplied: prepared.onCompactionApplied } : {}),
                     ...this.seamCCompactionOptions(prepared),
                     ...this.compactionHookOptions(spec, prepared.sessionId, "auto"),
@@ -3630,7 +3539,6 @@ function startTimeout(harness, abortController, remainingMs, softSuspendable = f
             abortController.abort();
             void harness.abort();
         }, hardMs);
-        timer.unref?.();
         state.clear = () => clearTimeout(timer);
     }
     return state;

package/dist/core/runner/session-rule-policy.js

@@ -1,10 +1,13 @@
 import { canonicalizeTarget, fileArgPath } from "../../tools/fs/safety.js";
+import { isWinFormPath } from "../../tools/fs/safety.js";
 import { createCoarseCommandNamePolicy } from "../tool-policy.js";
 import { canonicalToolName } from "../tool-name-aliases.js";
 export const PATH_WRITE_TOOLS = new Set(["Write", "Edit", "MultiEdit"]);
 export function isWithin(root, p) {
+    if (!root)
+        return false;
     const norm = (x) => {
-        let n = x.replace(/\\/g, "/");
+        let n = isWinFormPath(x) ? x.replace(/\\/g, "/") : x;
         if (/^[A-Za-z]:(\/|$)/.test(n))
             n = n[0].toLowerCase() + n.slice(1);
         return n;

package/dist/core/runner/teardown-bounded.d.ts

@@ -0,0 +1,7 @@
+export declare const TEARDOWN_LEG_MAX_MS = 15000;
+export declare function setTeardownLegBudgetForTests(ms: number | undefined): void;
+export declare class TeardownLegTimeoutError extends Error {
+    readonly leg: string;
+    constructor(leg: string, budgetMs: number);
+}
+export declare function settleTeardownLeg(leg: () => Promise<unknown> | unknown, label: string, report?: (err: Error) => void): Promise<void>;

package/dist/core/runner/teardown-bounded.js

@@ -0,0 +1,36 @@
+export const TEARDOWN_LEG_MAX_MS = 15_000;
+let teardownLegBudgetMs = TEARDOWN_LEG_MAX_MS;
+export function setTeardownLegBudgetForTests(ms) {
+    teardownLegBudgetMs = ms ?? TEARDOWN_LEG_MAX_MS;
+}
+export class TeardownLegTimeoutError extends Error {
+    leg;
+    constructor(leg, budgetMs) {
+        super(`teardown leg "${leg}" did not settle within ${budgetMs}ms — proceeding without it. ` +
+            `The leg keeps running detached; whatever it owns (process/env/store row) may be leaked until it settles on its own.`);
+        this.name = "TeardownLegTimeoutError";
+        this.leg = leg;
+    }
+}
+export async function settleTeardownLeg(leg, label, report) {
+    let timer;
+    try {
+        await Promise.race([
+            Promise.resolve().then(leg),
+            new Promise((_, reject) => {
+                timer = setTimeout(() => reject(new TeardownLegTimeoutError(label, teardownLegBudgetMs)), teardownLegBudgetMs);
+            }),
+        ]);
+    }
+    catch (e) {
+        try {
+            report?.(e instanceof Error ? e : new Error(String(e)));
+        }
+        catch {
+        }
+    }
+    finally {
+        if (timer !== undefined)
+            clearTimeout(timer);
+    }
+}

package/dist/core/runner/tool-disclosure.d.ts

@@ -2,6 +2,7 @@ import { type TSchema } from "typebox";
 import type { AgentMessage, AgentTool, AgentToolResult, AgentToolUpdateCallback, ToolExecutionMode } from "../../internal/harness-types.js";
 import type { Model } from "../../internal/llm.js";
 import type { ToolSpec } from "../types.js";
+import type { ToolFingerprintInput } from "../cache-break-detector.js";
 export declare const TOOL_SEARCH_NAME = "ToolSearch";
 export declare const TOOL_SEARCH_DEFAULT_MAX_RESULTS = 5;
 export interface DeferredToolInfo {
@@ -14,11 +15,7 @@ export declare function safeName(name: string): string;
 export declare function classifyDeferred(opts: {
     specs: ReadonlyArray<ToolSpec>;
     mcpToolNames: ReadonlyArray<string>;
-    fullTools: ReadonlyArray<{
-        name: string;
-        description: string;
-        parameters: unknown;
-    }>;
+    fullTools: ReadonlyArray<ToolFingerprintInput>;
     deferMode?: "auto";
     model: Model;
     deferNames?: ReadonlyArray<string>;