harnery 0.44.0 → 0.45.0

package/src/core/workflow/recovery-audit.ts

@@ -0,0 +1,181 @@
+import { randomUUID } from "node:crypto";
+import { existsSync, lstatSync, mkdirSync, readdirSync } from "node:fs";
+import { hostname } from "node:os";
+import { dirname, join } from "node:path";
+import { stateDirMode } from "../storage/modes.ts";
+import { assertWorkflowRunId } from "./approvals.ts";
+import {
+  fsyncParentDirectory,
+  readJsonRecord,
+  stableDigest,
+  writeImmutableJson,
+} from "./durable-record.ts";
+
+export interface WorkflowRecoveryAuthorization {
+  actor: string;
+  reason: string;
+}
+
+export interface WorkflowRecoveryIntent extends WorkflowRecoveryAuthorization {
+  schema_version: 1;
+  operation_id: string;
+  run_id: string;
+  action: "reclaim_unknown_liveness" | "unlock_resume";
+  target_sha256: string;
+  target: Record<string, unknown>;
+  recorded_at: string;
+  pid: number;
+  host: string;
+  outcome: "unknown";
+}
+
+export interface WorkflowRecoveryRecord {
+  intent: WorkflowRecoveryIntent;
+  outcome: "unknown" | "completed";
+}
+
+export function validateWorkflowRecoveryAuthorization(
+  authorization: WorkflowRecoveryAuthorization,
+): void {
+  for (const [field, limit] of [
+    ["actor", 200],
+    ["reason", 2_000],
+  ] as const) {
+    const value = authorization[field];
+    if (
+      typeof value !== "string" ||
+      !value.trim() ||
+      value.length > limit ||
+      Array.from(value).some(
+        (character) => character.charCodeAt(0) < 32 || character.charCodeAt(0) === 127,
+      )
+    ) {
+      throw new Error(
+        `workflow recovery requires a nonempty ${field} of at most ${limit} characters without control characters`,
+      );
+    }
+  }
+}
+
+/** Canonical recovery ledger: immutable intent is durable before effects.
+ * Completion cites that exact intent; absence always means unknown. */
+export function beginWorkflowRecovery(
+  coordRoot: string,
+  input: Omit<
+    WorkflowRecoveryIntent,
+    "schema_version" | "operation_id" | "recorded_at" | "pid" | "host" | "outcome"
+  >,
+): { intent: WorkflowRecoveryIntent; assertCurrent(): void; complete(): void } {
+  assertWorkflowRunId(input.run_id);
+  validateWorkflowRecoveryAuthorization(input);
+  if (!/^[a-f0-9]{64}$/.test(input.target_sha256))
+    throw new Error("workflow recovery requires an exact preview digest");
+  const intent: WorkflowRecoveryIntent = {
+    ...input,
+    schema_version: 1,
+    operation_id: `recovery-${randomUUID()}`,
+    recorded_at: new Date().toISOString(),
+    pid: process.pid,
+    host: hostname(),
+    outcome: "unknown",
+  };
+  const directory = join(
+    coordRoot,
+    ".harnery",
+    "workflows",
+    input.run_id,
+    "recovery",
+    intent.operation_id,
+  );
+  // Sync each newly published directory name too. Syncing only the intent's
+  // immediate parent cannot make a newly created ancestor durable.
+  const runDirectory = join(coordRoot, ".harnery", "workflows", input.run_id);
+  ensureDurableDirectory(runDirectory);
+  const ledgerDirectory = join(runDirectory, "recovery");
+  ensureDurableDirectory(ledgerDirectory);
+  fsyncParentDirectory(ledgerDirectory);
+  mkdirSync(directory, { mode: stateDirMode() });
+  fsyncParentDirectory(directory);
+  const intentPath = join(directory, "intent.json");
+  writeImmutableJson(intentPath, intent);
+  const assertCurrent = () => {
+    if (stableDigest(readJsonRecord(intentPath, "recovery intent")) !== stableDigest(intent))
+      throw new Error("workflow recovery audit authority changed");
+  };
+  return {
+    intent,
+    assertCurrent,
+    complete: () => {
+      assertCurrent();
+      writeImmutableJson(join(directory, "completion.json"), {
+        schema_version: 1,
+        operation_id: intent.operation_id,
+        intent_sha256: stableDigest(intent),
+        outcome: "completed",
+        recorded_at: new Date().toISOString(),
+      });
+    },
+  };
+}
+
+function ensureDurableDirectory(path: string): void {
+  if (existsSync(path)) {
+    if (!lstatSync(path).isDirectory())
+      throw new Error("workflow recovery ledger is not a directory");
+    return;
+  }
+  ensureDurableDirectory(dirname(path));
+  try {
+    mkdirSync(path, { mode: stateDirMode() });
+  } catch (error) {
+    if ((error as NodeJS.ErrnoException).code !== "EEXIST" || !lstatSync(path).isDirectory())
+      throw error;
+  }
+  fsyncParentDirectory(path);
+}
+
+export function readWorkflowRecoveryLedger(
+  coordRoot: string,
+  runId: string,
+): WorkflowRecoveryRecord[] {
+  assertWorkflowRunId(runId);
+  const directory = join(coordRoot, ".harnery", "workflows", runId, "recovery");
+  if (!existsSync(directory)) return [];
+  return readdirSync(directory)
+    .sort()
+    .map((entry) => {
+      if (!/^recovery-[a-f0-9-]{36}$/.test(entry))
+        throw new Error("invalid workflow recovery ledger entry");
+      const intent = readJsonRecord<WorkflowRecoveryIntent>(
+        join(directory, entry, "intent.json"),
+        "recovery intent",
+      );
+      if (
+        intent.schema_version !== 1 ||
+        intent.operation_id !== entry ||
+        intent.run_id !== runId ||
+        intent.outcome !== "unknown" ||
+        !["reclaim_unknown_liveness", "unlock_resume"].includes(intent.action) ||
+        !/^[a-f0-9]{64}$/.test(intent.target_sha256)
+      ) {
+        throw new Error("invalid workflow recovery intent authority");
+      }
+      validateWorkflowRecoveryAuthorization(intent);
+      const completionPath = join(directory, entry, "completion.json");
+      if (!existsSync(completionPath)) return { intent, outcome: "unknown" };
+      const completion = readJsonRecord<{
+        schema_version: number;
+        operation_id: string;
+        intent_sha256: string;
+        outcome: string;
+      }>(completionPath, "recovery completion");
+      if (
+        completion.schema_version !== 1 ||
+        completion.operation_id !== entry ||
+        completion.intent_sha256 !== stableDigest(intent) ||
+        completion.outcome !== "completed"
+      )
+        throw new Error("invalid workflow recovery completion authority");
+      return { intent, outcome: "completed" };
+    });
+}

package/src/core/workflow/run-state.ts

@@ -3,6 +3,8 @@ import {
   chmodSync,
   closeSync,
   existsSync,
+  fsyncSync,
+  lstatSync,
   mkdirSync,
   openSync,
   readFileSync,
@@ -23,7 +25,13 @@ import {
 import { stateFileMode } from "../storage/modes.ts";
 import { assertWorkflowRunId, readWorkflowApproval } from "./approvals.ts";
 import { isCanonicalWorkflowAttemptContext } from "./attempt-context.ts";
-import { readJsonRecord, writeImmutableJson } from "./durable-record.ts";
+import {
+  fsyncParentDirectory,
+  readJsonRecord,
+  stableDigest,
+  writeImmutableJson,
+} from "./durable-record.ts";
+import { beginWorkflowRecovery, type WorkflowRecoveryAuthorization } from "./recovery-audit.ts";
 import { normalizeWorkflowSpecialists } from "./specialists.ts";
 import type {
   WorkflowAttemptContext,
@@ -38,6 +46,7 @@ import type {
   WorkspaceBinding,
   WorkspaceCompatibilityExecutionEvidence,
 } from "./workspaces/index.ts";
+import { acquireNoClobberLease } from "./workspaces/leases.ts";
 import {
   assertWorkspaceManifestAuthority,
   isWorkspaceBinding,
@@ -204,8 +213,7 @@ export function workflowRunManifestPath(coordRoot: string, runId: string): strin
 }
 
 /** Acquire an exclusive lease shared by resume and reclaim. Existing locks
- * always refuse: a stale observation cannot authorize unlinking a new owner.
- * Crash recovery requires an audited ownership check outside this operation. */
+ * always refuse; explicit audited recovery is a separate operation. */
 export function acquireWorkflowResumeLease(coordRoot: string, runId: string): () => void {
   assertWorkflowRunId(runId);
   const path = join(coordRoot, ".harnery", "workflows", runId, "resume.lock");
@@ -218,23 +226,27 @@ export function acquireWorkflowResumeLease(coordRoot: string, runId: string): ()
   const ownerBytes = `${JSON.stringify(owner)}\n`;
   mkdirSync(dirname(path), { recursive: true });
 
-  const acquire = (): boolean => {
+  const acquire = (): { device: number; inode: number } | false => {
     try {
       const fd = openSync(path, "wx", stateFileMode());
       try {
         writeFileSync(fd, ownerBytes, "utf8");
+        fsyncSync(fd);
       } finally {
         closeSync(fd);
       }
       chmodSync(path, stateFileMode());
-      return true;
+      fsyncParentDirectory(path);
+      const identity = lstatSync(path);
+      return { device: identity.dev, inode: identity.ino };
     } catch (error) {
       if ((error as NodeJS.ErrnoException).code !== "EEXIST") throw error;
       return false;
     }
   };
 
-  if (!acquire()) {
+  const acquired = withResumeLeaseMutation(coordRoot, runId, acquire);
+  if (!acquired) {
     const existing = readResumeLease(path);
     const holder = existing
       ? ` by pid ${existing.pid} on ${existing.host}; its liveness has not been verified`
@@ -244,13 +256,129 @@ export function acquireWorkflowResumeLease(coordRoot: string, runId: string): ()
 
   return () => {
     try {
-      if (readFileSync(path, "utf8") === ownerBytes) unlinkSync(path);
+      // Only this live process can release this nonce. Acquisition never
+      // replaces a present lock, and unlock requires ESRCH for its owner, so
+      // neither supported writer can replace this lock during synchronous
+      // release. Taking their mutation gate would strand a live owner's lock
+      // on ordinary contention. Pin both identity and bytes for stale closures.
+      const identity = lstatSync(path);
+      if (
+        identity.isFile() &&
+        identity.dev === acquired.device &&
+        identity.ino === acquired.inode &&
+        readFileSync(path, "utf8") === ownerBytes
+      ) {
+        unlinkSync(path);
+        fsyncParentDirectory(path);
+      }
     } catch {
       // Leave an unreadable or replaced lock untouched.
     }
   };
 }
 
+export interface WorkflowResumeRecoveryOptions extends Partial<WorkflowRecoveryAuthorization> {
+  yes?: boolean;
+  lockSha256?: string;
+}
+
+export interface WorkflowResumeRecoveryResult {
+  action: "preview" | "unlocked";
+  run_id: string;
+  lock_sha256: string | null;
+  owner_state: "dead" | "live" | "unknown" | "absent";
+  owner: { pid: number; host: string; created_at: string } | null;
+  recovery_id?: string;
+}
+
+export function recoverWorkflowResumeLease(
+  coordRoot: string,
+  runId: string,
+  options: WorkflowResumeRecoveryOptions,
+): WorkflowResumeRecoveryResult {
+  assertWorkflowRunId(runId);
+  const path = join(coordRoot, ".harnery", "workflows", runId, "resume.lock");
+  const inspect = (): WorkflowResumeRecoveryResult => {
+    let lockSha256: string;
+    try {
+      const stat = lstatSync(path);
+      if (!stat.isFile() || stat.size > 8 * 1024)
+        throw new Error("resume lock is not a bounded regular file");
+      lockSha256 = stableDigest({
+        bytes: readFileSync(path, "utf8"),
+        device: stat.dev,
+        inode: stat.ino,
+      });
+    } catch (error) {
+      if ((error as NodeJS.ErrnoException).code !== "ENOENT") throw error;
+      return {
+        action: "preview",
+        run_id: runId,
+        lock_sha256: null,
+        owner_state: "absent",
+        owner: null,
+      };
+    }
+    const owner = readResumeLease(path);
+    let ownerState: WorkflowResumeRecoveryResult["owner_state"] = "unknown";
+    if (owner?.host === hostname()) {
+      try {
+        process.kill(owner.pid, 0);
+        ownerState = "live";
+      } catch (error) {
+        if ((error as NodeJS.ErrnoException).code === "ESRCH") ownerState = "dead";
+      }
+    }
+    return {
+      action: "preview",
+      run_id: runId,
+      lock_sha256: lockSha256,
+      owner_state: ownerState,
+      owner,
+    };
+  };
+  if (!options.yes) return inspect();
+  return withResumeLeaseMutation(coordRoot, runId, () => {
+    const observed = inspect();
+    if (observed.owner_state !== "dead")
+      throw new Error(`workflow resume unlock refused: owner is ${observed.owner_state}`);
+    if (!options.lockSha256 || options.lockSha256 !== observed.lock_sha256)
+      throw new Error("workflow resume unlock requires the exact current preview digest");
+    const audit = beginWorkflowRecovery(coordRoot, {
+      run_id: runId,
+      action: "unlock_resume",
+      target_sha256: options.lockSha256,
+      target: { lock: "resume.lock", owner: observed.owner },
+      actor: options.actor!,
+      reason: options.reason!,
+    });
+    const current = inspect();
+    if (current.lock_sha256 !== observed.lock_sha256 || current.owner_state !== "dead")
+      throw new Error("workflow resume lock authority changed before unlock");
+    audit.assertCurrent();
+    unlinkSync(path);
+    fsyncParentDirectory(path);
+    audit.complete();
+    return { ...observed, action: "unlocked", recovery_id: audit.intent.operation_id };
+  });
+}
+
+function withResumeLeaseMutation<T>(coordRoot: string, runId: string, action: () => T): T {
+  const authority = stableDigest({ run_id: runId, lock: "resume.lock" });
+  const lease = acquireNoClobberLease({
+    path: join(coordRoot, ".harnery", "workflows", runId, "resume-mutation.lease"),
+    scope: "workflow-resume-mutation",
+    authoritySha256: authority,
+    staleAfterMs: 300_000,
+    validateStaleOwner: (owner) => owner.authority_sha256 === authority,
+  });
+  try {
+    return action();
+  } finally {
+    lease.release();
+  }
+}
+
 function readResumeLease(path: string): { pid: number; host: string; created_at: string } | null {
   try {
     const value = JSON.parse(readFileSync(path, "utf8")) as Record<string, unknown>;
@@ -258,7 +386,8 @@ function readResumeLease(path: string): { pid: number; host: string; created_at:
       Number.isSafeInteger(value.pid) &&
       value.pid > 0 &&
       typeof value.host === "string" &&
-      typeof value.created_at === "string"
+      value.host.length > 0 &&
+      validTimestamp(value.created_at)
       ? { pid: value.pid, host: value.host, created_at: value.created_at }
       : null;
   } catch {

package/src/core/workflow/workspaces/local-git.ts

@@ -9,6 +9,7 @@ import {
 } from "node:fs";
 import { basename, dirname, isAbsolute, join, relative, resolve } from "node:path";
 import { stateDirMode, stateFileMode } from "../../storage/modes.ts";
+import { beginWorkflowRecovery, validateWorkflowRecoveryAuthorization } from "../recovery-audit.ts";
 import { acquireWorkflowResumeLease } from "../run-state.ts";
 import { WorkspaceAttestationError } from "./attestation-error.ts";
 import { cleanupWorkspace } from "./cleanup.ts";
@@ -53,7 +54,7 @@ import {
   type WorkspaceReclaimPreview,
   type WorkspaceReclaimResult,
 } from "./reclaim.ts";
-import { readWorkflowLiveness } from "./recovery.ts";
+import { hasActiveWorkflowEvidence, readWorkflowLiveness } from "./recovery.ts";
 import {
   appendWorkspaceEvent,
   fileSha256,
@@ -820,6 +821,7 @@ function reclaimPreview(
     run_id: binding.run_id,
     mode,
     liveness: readWorkflowLiveness(coordRoot, binding.run_id),
+    evidence_sha256: reclaimEvidenceDigest(coordRoot, binding),
     ownership: "unproven",
     dirty_paths: null,
     lease_holder: reclaimLeaseHolders(coordRoot, binding),
@@ -839,14 +841,45 @@ function reclaimPreview(
   return preview;
 }
 
-function assertReclaimLiveness(coordRoot: string, binding: WorkspaceBinding): void {
+function reclaimEvidenceDigest(coordRoot: string, binding: WorkspaceBinding): string {
+  const evidence = (name: string): string | null => {
+    try {
+      return readFileSync(join(coordRoot, ".harnery", "workflows", binding.run_id, name), "utf8");
+    } catch (error) {
+      if ((error as NodeJS.ErrnoException).code === "ENOENT") return null;
+      throw error;
+    }
+  };
+  return stableDigest({
+    binding,
+    proof: evidence("proof.json"),
+    transcript: evidence("transcript.jsonl"),
+  });
+}
+
+function assertReclaimLiveness(
+  coordRoot: string,
+  binding: WorkspaceBinding,
+  recovery?: WorkspaceReclaimOptions["unknownLiveness"],
+): void {
   const liveness = readWorkflowLiveness(coordRoot, binding.run_id);
+  if (liveness === "active" || (recovery && hasActiveWorkflowEvidence(coordRoot, binding.run_id))) {
+    throw new WorkflowReclaimError(
+      "workflow_reclaim_run_live",
+      `run ${binding.run_id} has active evidence; its workspace was left untouched`,
+    );
+  }
+  if (recovery && reclaimEvidenceDigest(coordRoot, binding) !== recovery.evidenceSha256) {
+    throw new WorkflowReclaimError(
+      "workflow_reclaim_recovery_mismatch",
+      "workflow reclaim requires the exact current preview evidence digest",
+    );
+  }
   if (liveness === "inactive") return;
+  if (liveness === "unknown" && recovery) return;
   throw new WorkflowReclaimError(
-    liveness === "active" ? "workflow_reclaim_run_live" : "workflow_reclaim_liveness_unknown",
-    liveness === "active"
-      ? `run ${binding.run_id} is active; its workspace was left untouched`
-      : `run ${binding.run_id} has missing, corrupt, or ambiguous liveness evidence; its workspace was left untouched`,
+    "workflow_reclaim_liveness_unknown",
+    `run ${binding.run_id} has missing, corrupt, or ambiguous liveness evidence; its workspace was left untouched`,
   );
 }
 
@@ -892,10 +925,11 @@ function reclaimAuthorityGuard(
   binding: WorkspaceBinding,
   claim: WorkspaceClaim,
   reservation: WorkspaceReclaimReservation,
+  recovery?: WorkspaceReclaimOptions["unknownLiveness"],
 ): void {
   try {
     reservation.assertCurrent();
-    assertReclaimLiveness(coordRoot, binding);
+    assertReclaimLiveness(coordRoot, binding, recovery);
     assertReclaimOwnership(coordRoot, binding);
     assertPeerWorkspaceAuthority(coordRoot, claim);
   } catch (error) {
@@ -914,10 +948,11 @@ async function reclaim(
     throw new Error("unsupported workspace reclaim mode");
   }
   if (!options.yes) return reclaimPreview(coordRoot, binding, options.mode);
+  if (options.unknownLiveness) validateWorkflowRecoveryAuthorization(options.unknownLiveness);
 
   // Liveness comes before terminal-proof cleanup helpers. A live run may not
   // have a proof yet, and a missing proof must never be discovered after reset.
-  assertReclaimLiveness(coordRoot, binding);
+  assertReclaimLiveness(coordRoot, binding, options.unknownLiveness);
   const claim = assertReclaimOwnership(coordRoot, binding);
   let reservation: WorkspaceReclaimReservation;
   try {
@@ -926,7 +961,7 @@ async function reclaim(
     throw new WorkflowReclaimError("workflow_reclaim_lease_held", (error as Error).message);
   }
   try {
-    reclaimAuthorityGuard(coordRoot, binding, claim, reservation);
+    reclaimAuthorityGuard(coordRoot, binding, claim, reservation, options.unknownLiveness);
     let releaseResume: () => void;
     try {
       releaseResume = acquireWorkflowResumeLease(coordRoot, binding.run_id);
@@ -953,9 +988,24 @@ async function reclaim(
           throw new WorkflowReclaimError("workflow_reclaim_lease_held", (error as Error).message);
         }
         try {
-          assertReclaimLiveness(coordRoot, binding);
+          assertReclaimLiveness(coordRoot, binding, options.unknownLiveness);
           assertReclaimOwnership(coordRoot, binding);
           const preview = reclaimPreview(coordRoot, binding, options.mode);
+          const audit = options.unknownLiveness
+            ? beginWorkflowRecovery(coordRoot, {
+                run_id: binding.run_id,
+                action: "reclaim_unknown_liveness",
+                actor: options.unknownLiveness.actor,
+                reason: options.unknownLiveness.reason,
+                target_sha256: options.unknownLiveness.evidenceSha256,
+                target: {
+                  binding_id: binding.binding_id,
+                  workspace_root: binding.workspace_root,
+                  mode: options.mode,
+                  liveness: preview.liveness,
+                },
+              })
+            : undefined;
           let preparation: ReclaimPreparation | undefined;
           // Keep ordinary cleanup's terminal-evidence checks, intent and receipts.
           // Its provider callback uses the leases already held here instead of
@@ -967,12 +1017,28 @@ async function reclaim(
             provider: {
               ...provider,
               cleanup: async (candidate, intent) => {
-                assertReclaimLiveness(coordRoot, candidate);
+                assertReclaimLiveness(coordRoot, candidate, options.unknownLiveness);
                 assertReclaimOwnership(coordRoot, candidate);
                 return cleanupUnderLease(coordRoot, candidate, claim, intent, {
                   mode: options.mode,
-                  beforeEffect: () =>
-                    reclaimAuthorityGuard(coordRoot, candidate, claim, reservation),
+                  recovery: options.unknownLiveness,
+                  beforeEffect: () => {
+                    audit?.assertCurrent();
+                    reclaimAuthorityGuard(
+                      coordRoot,
+                      candidate,
+                      claim,
+                      reservation,
+                      options.unknownLiveness,
+                    );
+                  },
+                  beforeRefEffect: () => {
+                    audit?.assertCurrent();
+                    reservation.assertCurrent();
+                    assertReclaimLiveness(coordRoot, candidate, options.unknownLiveness);
+                    requiredClaim(coordRoot, candidate);
+                    assertPeerWorkspaceAuthority(coordRoot, claim);
+                  },
                   prepared: (value) => {
                     preparation = value;
                   },
@@ -991,6 +1057,8 @@ async function reclaim(
                 "workspace cleanup preconditions could not be proved",
             );
           }
+          if (cleanupResult.status === "released" || cleanupResult.status === "already_released")
+            audit?.complete();
           return {
             ...preview,
             action: "reclaimed",
@@ -1104,7 +1172,9 @@ function cleanupUnderLease(
   intent: WorkspaceCleanupIntent,
   reclaimRequest?: {
     mode: ReclaimMode;
+    recovery?: WorkspaceReclaimOptions["unknownLiveness"];
     beforeEffect: () => void;
+    beforeRefEffect: () => void;
     prepared: (value: ReclaimPreparation) => void;
   },
   beforeRemoval?: () => void,
@@ -1152,7 +1222,7 @@ function cleanupUnderLease(
     };
   }
   const liveness = readWorkflowLiveness(coordRoot, binding.run_id);
-  if (liveness !== "inactive") {
+  if (liveness !== "inactive" && !(liveness === "unknown" && reclaimRequest?.recovery)) {
     return {
       ...base,
       status: "blocked",
@@ -1245,7 +1315,7 @@ function cleanupUnderLease(
   }
 
   if (reclaimRequest) {
-    assertReclaimLiveness(coordRoot, binding);
+    assertReclaimLiveness(coordRoot, binding, reclaimRequest.recovery);
     assertReclaimOwnership(coordRoot, binding);
     revalidateClaimResources(claim, root);
     reclaimRequest.prepared(
@@ -1293,6 +1363,7 @@ function cleanupUnderLease(
       return { ...base, status: "blocked", reason: "workspace ref changed during cleanup" };
     }
     revalidateClaimResources(claim, root);
+    reclaimRequest?.beforeRefEffect();
     const deleted = gitMaybe(repository.source_root.realpath, [
       "update-ref",
       "-d",

package/src/core/workflow/workspaces/reclaim.ts

@@ -16,6 +16,7 @@
  */
 
 import { existsSync } from "node:fs";
+import type { WorkflowRecoveryAuthorization } from "../recovery-audit.ts";
 import { git, gitMaybe } from "./git.ts";
 import type { WorkflowLiveness } from "./recovery.ts";
 import type { WorkspaceCleanupAttempt, WorkspaceCleanupReceipt } from "./types.ts";
@@ -25,6 +26,7 @@ export type ReclaimMode = "salvage" | "discard";
 export type WorkflowReclaimRefusalCode =
   | "workflow_reclaim_run_live"
   | "workflow_reclaim_liveness_unknown"
+  | "workflow_reclaim_recovery_mismatch"
   | "workflow_reclaim_ownership_unproven"
   | "workflow_reclaim_lease_held";
 
@@ -41,6 +43,7 @@ export class WorkflowReclaimError extends Error {
 export interface WorkspaceReclaimOptions {
   mode: ReclaimMode;
   yes?: boolean;
+  unknownLiveness?: WorkflowRecoveryAuthorization & { evidenceSha256: string };
 }
 
 export interface ReclaimLeaseHolder {
@@ -56,6 +59,7 @@ export interface WorkspaceReclaimPreview {
   run_id: string;
   mode: ReclaimMode;
   liveness: WorkflowLiveness;
+  evidence_sha256: string;
   ownership: "ok" | "unproven";
   ownership_detail?: string;
   dirty_paths: string[] | null;

package/src/core/workflow/workspaces/recovery.ts

@@ -4,6 +4,29 @@ import { readWorkflowProof } from "../proof.ts";
 
 export type WorkflowLiveness = "active" | "inactive" | "unknown";
 
+/** Corruption must not hide a readable active marker from an audited override. */
+export function hasActiveWorkflowEvidence(coordRoot: string, runId: string): boolean {
+  const path = join(resolve(coordRoot), ".harnery", "workflows", runId, "transcript.jsonl");
+  if (!existsSync(path)) return false;
+  let active = false;
+  for (const line of readFileSync(path, "utf8").split("\n")) {
+    try {
+      const event = JSON.parse(line);
+      if (event?.run_id !== runId) continue;
+      if (["run.start", "run.resume", "agent.start"].includes(event.event)) active = true;
+      if (
+        event.event === "run.parked" ||
+        (event.event === "run.end" && typeof event.ok === "boolean") ||
+        (event.event === "workspace.reattach.failed" && ["blocked", "lost"].includes(event.status))
+      )
+        active = false;
+    } catch {
+      /* Unreadable evidence cannot establish inactivity. */
+    }
+  }
+  return active;
+}
+
 export function readWorkflowLiveness(coordRoot: string, runId: string): WorkflowLiveness {
   const root = resolve(coordRoot);
   const transcriptPath = join(root, ".harnery", "workflows", runId, "transcript.jsonl");