fix(pi-workflow): reset failed verification on retry

extensions/pi-workflow/engine.test.ts

@@ -1512,9 +1512,20 @@ test("Recovery retry reruns failed verification after the remediation limit", as
   await withTempDir(async (directory) => {
     const setup = await fixture(directory);
     let commandRuns = 0;
+    let runId: string | undefined;
     const hostCommands: HostCommandExecutor = {
       async run() {
         commandRuns++;
+        if (commandRuns === 2) {
+          assert.ok(runId);
+          const retrying = await setup.store.loadRun(runId);
+          const [command] = retrying.verificationEvidence?.commands ?? [];
+          assert.equal(command?.status, "running");
+          assert.equal(command?.stdout, undefined);
+          assert.equal(command?.stderr, undefined);
+          assert.equal(command?.exitCode, undefined);
+          assert.equal(command?.error, undefined);
+        }
         return {
           result: {
             code: commandRuns === 1 ? 1 : 0,
@@ -1552,6 +1563,7 @@ test("Recovery retry reruns failed verification after the remediation limit", as
       hostCommands,
       effectiveConfig,
     }).execute(startCommand());
+    runId = planned.runId;
     const approved = await setup.engine("session-1", {
       hostCommands,
       effectiveConfig,
@@ -1598,6 +1610,7 @@ test("Recovery retry reruns failed verification after the remediation limit", as
     const liveStolen = structuredClone(untouched);
     liveStolen.revision++;
     liveStolen.parentSessionIds.push("transient-session");
+    liveStolen.verificationEvidence!.commands[0]!.error = "stale host exception";
     liveStolen.lease = {
       processId: 200,
       activeParentSessionId: "transient-session",

extensions/pi-workflow/engine.ts

@@ -4279,6 +4279,10 @@ export class WorkflowEngine {
         run.verificationEvidence?.commands.filter(
           (command) => command.status === "running" || command.status === "interrupted",
         ) ?? [];
+      const failedCommands =
+        run.verificationEvidence?.commands.filter(
+          (command) => command.status === "failed",
+        ) ?? [];
       run = await this.transition(run, (next) => {
         for (const assignment of next.assignments) {
           if (
@@ -4309,6 +4313,21 @@ export class WorkflowEngine {
           next.script.error = undefined;
         }
         if (next.verificationEvidence) {
+          for (const command of failedCommands) {
+            const persisted = next.verificationEvidence.commands.find(
+              (candidate) => candidate.id === command.id,
+            );
+            if (!persisted) continue;
+            persisted.status = "pending";
+            persisted.exitCode = undefined;
+            persisted.stdout = undefined;
+            persisted.stderr = undefined;
+            persisted.error = undefined;
+          }
+          if (failedCommands.length > 0) {
+            next.verificationEvidence.focusedBehavioralTests = false;
+            next.verificationEvidence.verifierSucceeded = false;
+          }
           for (const command of interruptedCommands) {
             const persisted = next.verificationEvidence.commands.find(
               (candidate) => candidate.id === command.id,