feat: block progress claims without execution evidence

hooks/force-recall/handler.ts

@@ -109,6 +109,9 @@ function buildGateLockInput(wrapperResult: any): Record<string, unknown> {
     && typeof wrapperResult.taskRecord?.task_name === "string"
     && wrapperResult.taskRecord.task_name.trim().length > 0;
   const hasButtonPathClosureEvidence = needsOwnerDecision && wrapperResult.silentLaunchOk === true;
+  const claimedProgression = wrapperResult.classification === "long_task"
+    ? "already progressing to the next step in background"
+    : "";
 
   return {
     classification: wrapperResult.classification,
@@ -118,6 +121,14 @@ function buildGateLockInput(wrapperResult: any): Record<string, unknown> {
     nextStep: hasConcreteExecutionEvidence ? requiredNextAction : "",
     requiredNextAction: hasConcreteExecutionEvidence ? requiredNextAction : "",
     concreteNextAction: hasConcreteExecutionEvidence ? requiredNextAction : "",
+    progressionClaim: claimedProgression,
+    claimedProgression: claimedProgression,
+    statusSummary: claimedProgression,
+    toolCallEvidence: "",
+    dispatchEvidence: "",
+    fileChangeEvidence: "",
+    verificationEvidence: "",
+    checkpointArtifactEvidence: hasExternalizedCheckpointEvidence ? wrapperResult.taskRecord.task_name.trim() : "",
     externalizedCheckpointPath: hasExternalizedCheckpointEvidence ? wrapperResult.taskRecord.task_name.trim() : "",
     externalizedTrigger: hasExternalizedCheckpointEvidence ? "hook-preflight-checkpoint" : "",
     handoffMode: hasButtonPathClosureEvidence ? (wrapperResult.handoff?.mode ?? "button_path") : "direct_reply",

scripts/long_task_gate_lock.mjs

@@ -17,6 +17,18 @@ const EVIDENCE_FIELDS = Object.freeze({
     'handoff.mode',
     'replyClosureMode',
   ]),
+  progressionClaim: Object.freeze([
+    'progressionClaim',
+    'claimedProgression',
+    'statusSummary',
+  ]),
+  executionEvidence: Object.freeze([
+    'toolCallEvidence',
+    'dispatchEvidence',
+    'fileChangeEvidence',
+    'verificationEvidence',
+    'checkpointArtifactEvidence',
+  ]),
 });
 
 const GATE_REQUIREMENTS = Object.freeze({
@@ -35,6 +47,11 @@ const GATE_REQUIREMENTS = Object.freeze({
     acceptedFields: EVIDENCE_FIELDS.buttonPathMode,
     requiredValue: 'button_path',
   }),
+  executionEvidence: Object.freeze({
+    evidenceKey: 'executionEvidence',
+    acceptedFields: EVIDENCE_FIELDS.executionEvidence,
+    requiredValue: 'tool call, dispatch, file change, verification output, or checkpoint artifact evidence',
+  }),
 });
 
 function fail(code, message) {
@@ -136,6 +153,16 @@ function usesButtonPath(input) {
   return hasAcceptedValue(input, EVIDENCE_FIELDS.buttonPathMode, 'button_path');
 }
 
+function claimsProgressionWithoutEvidence(input) {
+  const progressionClaim = EVIDENCE_FIELDS.progressionClaim
+    .map((fieldPath) => getPathValue(input, fieldPath))
+    .find((value) => hasNonEmptyString(value));
+
+  if (!hasNonEmptyString(progressionClaim)) return false;
+
+  return !hasAnyNonEmptyString(input, EVIDENCE_FIELDS.executionEvidence);
+}
+
 function evaluateGate(input) {
   const gateRequired = isLongTask(input);
   const reasons = [];
@@ -175,6 +202,13 @@ function evaluateGate(input) {
     allowedResponseModes.push('button_path');
   }
 
+  if (claimsProgressionWithoutEvidence(input)) {
+    failed = true;
+    reasons.push('claimed progression without concrete execution evidence is forbidden');
+    requiredEvidence.push(describeRequirement(GATE_REQUIREMENTS.executionEvidence));
+    allowedResponseModes.push('evidence_preserving_follow_up');
+  }
+
   if (!failed) {
     reasons.push('required long-task gate evidence is present or no gated condition was triggered');
     allowedResponseModes.push(needsOwnerDecision(input) ? 'button_path' : 'direct_reply');