@ornncompute/cli 0.1.7 → 0.1.8

package/README.md

@@ -56,10 +56,11 @@ ornn nodes keys list <node-id> [--json]
 ornn node health <node-id> [--json]
 ornn node diagnose <node-id> [--json]
 ornn node deenroll <node-id> [--reason <text>] [--keep-record] [--json]
-ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... [--ssh-user ubuntu|admin|ornn] [--json]
-ornn fleet clean <failed-fleet-id> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... [--json]
-ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash> [--source-user <user>]... [--preserve-user <user>]... [--json]
+ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] [--ssh-user ubuntu|admin|ornn] [--json]
+ornn fleet clean <fleet-id> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] [--json]
+ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash> [--json]
 ornn fleet enroll <fleet-id> --identity-file <path> [--confirm-takeover <ip[,ip...]>] [--json]
+ornn fleet delete <fleet-id> --confirm-delete <fleet-id> [--json]
 ornn reservations commerce list --tenant <email-or-id> [--fleet <fleet-id>] [--json]
 ornn reservations commerce create --tenant <email> --listing <listing-id> --fleet <fleet-id> [--start-at <timestamp>] [--end-at <timestamp>] [--price-per-gpu-hour <rate>] [--json]
 ornn fleet deploy <fleet-id> --tenant <contact-email> [--user <email-or-id>] --commerce-reservation <id> [--network public|private] [--json]
@@ -227,28 +228,56 @@ Fleet handoff is an ordered, Compute-backed workflow:
 - Before first contact, independently verify each host key and add it to the
   operator's SSH `known_hosts`; fleet SSH uses strict host-key checking and will
   not trust a key first presented during cleanup.
-- `fleet clean ... --dry-run` tries `ubuntu`, `admin`, then `ornn`, records each
-  host's hardware identity and exact deletion plan, and prints one fleet plan
-  hash. Repeat `--source-user` for Linux accounts owned by the departing tenant
-  (including an explicitly reviewed orphaned `/home/<user>` directory) and
-  `--preserve-user` for reviewed host accounts. Package/service-owned
-  accounts are protected automatically; any other account stops planning for
-  explicit review. Re-running with `--confirm-clean <plan-hash>` and the same
-  account flags removes the approved tenant
-  users, keys, containers/images/volumes/networks/build cache, workload paths,
-  mounts, services, and tenant-scoped logs. Apply refuses plans with running
-  containers, locks the management/root passwords, enforces key-only SSH, trims
-  supported filesystems, preserves existing root and management keys and host
-  logs, and stores per-node proof in Compute. A failed
-  cleanup can be dry-run again with the same fleet ID; only failed members are
-  replanned. Residual GPU
-  consumers or tenant/platform listeners fail verification instead of being
-  silently accepted.
+- `fleet clean ... --dry-run` tries `ubuntu`, `admin`, then `ornn`, binds the
+  inventory to the host's hardware identity, records the plan in Compute, and
+  writes an editable mode-`0600` policy beside the local fleet manifest (or to
+  `--policy-out`). Inventory covers accounts and orphan homes, authorized SSH
+  keys, tenant/log/temp/platform paths, workload and listener processes, GPU
+  consumers, custom systemd services, Docker/Podman objects and bind-mount
+  dependencies, and every block drive. Each resource advertises its own valid
+  actions. Suggestions are never applied silently: unresolved choices and
+  conflicts are recorded as plan data, so edit the policy and re-run the same
+  fleet with `--policy`; planning is not terminalized just because a choice is
+  needed. `--source-user` and `--preserve-user` remain legacy user-only
+  shortcuts and cannot be mixed with a policy file.
+- `fleet clean <fleet-id> ... --confirm-clean <plan-hash>` is shown only after
+  every choice is resolved. Execution rechecks hardware and resource identity,
+  runs an idempotent per-resource journal, preserves the management SSH key,
+  enforces key-only SSH, and returns exact deleted, preserved, exception, and
+  drive-assurance evidence. `ignore` requires a reason and produces
+  `clean_with_exceptions`; it is never reported as residue-free deletion.
+  Data-drive actions are selectable: `logical_clean`, `discard`, `sanitize`, or
+  `reformat`. The system drive and drives without a unique serial/WWN are
+  locked to keep/ignore for in-band cleanup. `sanitize` requires NVMe plus
+  controller-reported completion. Logical cleanup, discard, and reformat are
+  not certified physical-media erasure, and the system drive requires an
+  out-of-band reimage for that assurance.
+  If the local SSH command times out while a long device operation is still in
+  progress, the recorded run remains resumable; re-run the same
+  `--confirm-clean <plan-hash>` command instead of creating a different plan.
+  Plans created by a pre-v6 runner must be replanned with `--dry-run` before
+  approval. This keeps newly discovered resources bound to a fresh reviewed
+  hash instead of adapting a legacy account-only plan during execution.
+
+In the generated policy, change only `action`, plus `reason` for `ignore` or
+`replacement` for `replace`; keep each generated `identity_hash` intact. `keep`
+leaves a resource running, `stop_keep` stops but retains it, `stop` terminates a
+process, `delete` removes it, `restart` restarts a service, and `replace`
+recreates a container from the supplied image/name/network/command/labels/mounts
+specification. The generated `suggested_action` is review guidance, not an
+automatic decision. Do not put secrets in replacement commands or labels; use
+the platform secret manager or a reviewed secret mount.
+
 - `fleet enroll <fleet-id>` refuses nodes without successful cleanup proof,
   issues and revokes one cleanup-bound token per host, installs the agent,
   assigns the IB island, verifies the same management account still has
   key-based SSH and passwordless sudo, and records the resulting GPU node IDs
   in Compute. Generic enrollment tokens retain the legacy enrollment sanitizer.
+- `fleet delete <fleet-id> --confirm-delete <fleet-id>` hard-deletes only an
+  abandoned cleanup plan or an unassigned legacy enrollment record. It saves a
+  local audit manifest first and refuses any Commerce/tenant/deployment link,
+  active enrollment, assigned GPU node, or completed/running cleanup evidence.
+  It never deletes the linked GPU-node records.
 - `reservations commerce create --tenant <email> --listing <listing-id>
 --fleet <fleet-id>` creates and activates the tenant's Commerce reservation.
   GPU count comes from the enrolled fleet; start, end, and hourly rate default

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ornncompute/cli",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "Command-line interface for Ornn compute access workflows.",
   "type": "module",
   "bin": {

package/src/cli.mjs

@@ -1,11 +1,11 @@
 import { spawn } from "node:child_process";
 import { randomUUID } from "node:crypto";
 import { createReadStream } from "node:fs";
-import { mkdir, mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
+import { chmod, mkdir, mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
 import { createRequire } from "node:module";
 import { isIP } from "node:net";
 import { homedir, tmpdir } from "node:os";
-import { basename, join } from "node:path";
+import { basename, dirname, join } from "node:path";
 
 import {
   CliApiError,
@@ -132,11 +132,12 @@ Usage:
   ornn tokens list [--json]
   ornn tokens create --operator <id> [--facility <id>] [--expires-in <seconds>] [--mode bare-metal|vm] [--ip <addr>] [--force] [--json]
   ornn tokens revoke <token-id> [--json]
-  ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... [--ssh-user ubuntu|admin|ornn] [--json]
-  ornn fleet clean <failed-fleet-id> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... [--json]
-  ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash> [--source-user <user>]... [--preserve-user <user>]... [--json]
+  ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] [--ssh-user ubuntu|admin|ornn] [--json]
+  ornn fleet clean <fleet-id> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] [--json]
+  ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash> [--json]
   ornn fleet enroll <fleet-id> --identity-file <path> [--confirm-takeover <ip[,ip...]>] [--json]
   ornn fleet deploy <fleet-id> --tenant <email> [--user <email-or-id>] --commerce-reservation <id> [--network public|private] [--json]
+  ornn fleet delete <fleet-id> --confirm-delete <fleet-id> [--json]
   ornn reservations withdraw <reservation-id> [--json]
   ornn reservations transfer <reservation-id> --target-tenant <id> [--target-user <id>] [--node <id>] [--strategy reject|park] [--confirm] [--json]
   ornn reservations commerce list --tenant <email-or-id> [--fleet <fleet-id>] [--json]
@@ -645,7 +646,12 @@ function validateHelpInvocation(command, args) {
         "--ib-island",
         "--identity-file",
         "--confirm-clean",
+        "--confirm-delete",
         "--confirm-takeover",
+        "--policy",
+        "--policy-out",
+        "--source-user",
+        "--preserve-user",
         "--ssh-user",
         "--ssh-port",
         "--parallel",
@@ -675,7 +681,12 @@ function validateHelpInvocation(command, args) {
         ? null
         : "Usage: ornn fleet deploy <fleet-id> --tenant <email> [--user <email-or-id>] --commerce-reservation <id>";
     }
-    return "Usage: ornn fleet clean|enroll|deploy";
+    if (subcommand === "delete") {
+      return positionals.length === 1
+        ? null
+        : "Usage: ornn fleet delete <fleet-id> --confirm-delete <fleet-id>";
+    }
+    return "Usage: ornn fleet clean|enroll|deploy|delete";
   }
 
   if (command === "nodes") {
@@ -2351,6 +2362,55 @@ function fleetAccountOptions(value, optionName) {
   return [...new Set(normalized)].sort();
 }
 
+async function readFleetCleanupPolicy(pathValue) {
+  if (!optionProvided(pathValue)) return null;
+  const path = expandUserPath(optionalStringOption(pathValue));
+  let policy;
+  try {
+    policy = JSON.parse(await readFile(path, "utf8"));
+  } catch (error) {
+    throw new Error(`Could not read cleanup policy ${path}: ${fleetResultError(error)}`);
+  }
+  if (!policy || typeof policy !== "object" || Array.isArray(policy)) {
+    throw new Error("Cleanup policy must be a JSON object.");
+  }
+  if (policy.schema_version !== 1) {
+    throw new Error("Cleanup policy schema_version must be 1.");
+  }
+  const direct =
+    policy.resources && typeof policy.resources === "object" && !Array.isArray(policy.resources);
+  const perNode = policy.nodes && typeof policy.nodes === "object" && !Array.isArray(policy.nodes);
+  if (Boolean(direct) === Boolean(perNode)) {
+    throw new Error("Cleanup policy must contain exactly one of resources or nodes.");
+  }
+  return { path, policy };
+}
+
+function cleanupPolicyForNode(loaded, node, nodeCount) {
+  if (!loaded) return null;
+  if (loaded.policy.resources) {
+    if (nodeCount !== 1) {
+      throw new Error(
+        "A direct resources policy can only be used for one node; use nodes keyed by IP for a fleet.",
+      );
+    }
+    return loaded.policy;
+  }
+  const policy = loaded.policy.nodes[node.ip_address] || loaded.policy.nodes[node.id];
+  if (!policy) {
+    throw new Error(`Cleanup policy has no entry for ${node.ip_address}.`);
+  }
+  if (
+    policy.schema_version !== 1 ||
+    !policy.resources ||
+    typeof policy.resources !== "object" ||
+    Array.isArray(policy.resources)
+  ) {
+    throw new Error(`Cleanup policy entry for ${node.ip_address} is invalid.`);
+  }
+  return policy;
+}
+
 async function fleet(args, context) {
   const [subcommand, ...rest] = args;
   if (subcommand === "clean") {
@@ -2362,12 +2422,75 @@ async function fleet(args, context) {
   if (subcommand === "deploy") {
     return await fleetDeploy(rest, context);
   }
-  throw new Error("Usage: ornn fleet clean|enroll|deploy");
+  if (subcommand === "delete") {
+    return await fleetDelete(rest, context);
+  }
+  throw new Error("Usage: ornn fleet clean|enroll|deploy|delete");
+}
+
+async function fleetDelete(args, context) {
+  const usage = "Usage: ornn fleet delete <fleet-id> --confirm-delete <fleet-id> [--json]";
+  const { options, positionals } = parseOptions(args, {
+    boolean: ["json"],
+    value: ["confirm-delete"],
+  });
+  if (positionals.length !== 1) {
+    throw new Error(usage);
+  }
+  const fleetId = positionals[0];
+  if (!isUuid(fleetId)) {
+    throw new Error("<fleet-id> must be a UUID.");
+  }
+  const confirmation = requiredOption(options.confirmDelete, "--confirm-delete");
+  if (confirmation !== fleetId) {
+    throw new Error("--confirm-delete must exactly match <fleet-id>.");
+  }
+
+  const preview = await operatorRequest({
+    endpoint: `/internal/fleets/${encodeURIComponent(fleetId)}`,
+    env: context.env,
+    fetchImpl: context.fetchImpl,
+  });
+  if (String(preview?.id || "") !== fleetId) {
+    throw new Error("Fleet lookup returned a different fleet id; deletion stopped.");
+  }
+  const manifestPath = await saveFleetManifest(
+    {
+      ...preview,
+      deletion: { requested_at: new Date().toISOString(), status: "requested" },
+    },
+    context.env,
+  );
+  const result = await operatorRequest({
+    body: { confirm_fleet_id: confirmation },
+    endpoint: `/internal/fleets/${encodeURIComponent(fleetId)}`,
+    env: context.env,
+    fetchImpl: context.fetchImpl,
+    method: "DELETE",
+  });
+  const archived = {
+    ...(result?.fleet || preview),
+    deletion: {
+      deleted_at: result?.deleted_at || new Date().toISOString(),
+      status: "deleted",
+    },
+  };
+  await saveFleetManifest(archived, context.env);
+  if (options.json) {
+    writeJson(context.stdout, { ...result, audit_manifest: manifestPath });
+  } else {
+    context.stdout.write(`Deleted fleet ${fleetId}.\n`);
+    context.stdout.write(
+      `Released members: ${Array.isArray(archived.nodes) ? archived.nodes.length : 0}\n`,
+    );
+    context.stdout.write(`Audit manifest: ${manifestPath}\n`);
+  }
+  return 0;
 }
 
 async function fleetClean(args, context) {
   const usage =
-    "Usage: ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... [--ssh-user ubuntu|admin|ornn] OR ornn fleet clean <failed-fleet-id> --identity-file <path> --dry-run [--source-user <user>]... [--preserve-user <user>]... OR ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash> [--source-user <user>]... [--preserve-user <user>]...";
+    "Usage: ornn fleet clean <ip>... --operator <id-or-slug> --ib-island <name> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] [--ssh-user ubuntu|admin|ornn] OR ornn fleet clean <fleet-id> --identity-file <path> --dry-run [--policy <path>] [--policy-out <path>] OR ornn fleet clean <fleet-id> --identity-file <path> --confirm-clean <plan-hash>";
   const { options, positionals } = parseOptions(args, {
     boolean: ["dry-run", "json"],
     value: [
@@ -2375,6 +2498,8 @@ async function fleetClean(args, context) {
       "ib-island",
       "identity-file",
       "confirm-clean",
+      "policy",
+      "policy-out",
       "source-user",
       "preserve-user",
       "ssh-user",
@@ -2402,30 +2527,57 @@ async function fleetClean(args, context) {
     : FLEET_DEFAULT_TIMEOUT_SECONDS;
   const sourceUsers = fleetAccountOptions(options.sourceUser, "--source-user");
   const preserveUsers = fleetAccountOptions(options.preserveUser, "--preserve-user");
+  const loadedPolicy = await readFleetCleanupPolicy(options.policy);
+  const policyOut = optionProvided(options.policyOut)
+    ? expandUserPath(optionalStringOption(options.policyOut))
+    : null;
   const accountOverlap = sourceUsers.filter((user) => preserveUsers.includes(user));
   if (accountOverlap.length) {
     throw new Error(`Users cannot be both source and preserved: ${accountOverlap.join(", ")}.`);
   }
+  if (loadedPolicy && (sourceUsers.length || preserveUsers.length)) {
+    throw new Error("Use either --policy or the legacy user shortcuts, not both.");
+  }
   const confirmHash = optionalStringOption(options.confirmClean);
   if (confirmHash) {
-    if (positionals.length !== 1 || options.dryRun || options.operator || options.ibIsland) {
+    if (
+      positionals.length !== 1 ||
+      options.dryRun ||
+      options.operator ||
+      options.ibIsland ||
+      loadedPolicy ||
+      policyOut ||
+      sourceUsers.length ||
+      preserveUsers.length
+    ) {
       throw new Error(usage);
     }
     if (!/^[0-9a-f]{64}$/.test(confirmHash)) {
       throw new Error("--confirm-clean must be the exact 64-character cleanup plan hash.");
     }
     const fleetId = positionals[0];
-    const approved = await operatorRequest({
-      body: {
-        plan_hash: confirmHash,
-        source_users: sourceUsers,
-        preserve_users: preserveUsers,
-      },
-      endpoint: `/internal/fleets/${encodeURIComponent(fleetId)}/cleanup/approve`,
-      env: context.env,
-      fetchImpl: context.fetchImpl,
-      method: "POST",
-    });
+    let approved;
+    try {
+      approved = await operatorRequest({
+        body: {
+          plan_hash: confirmHash,
+        },
+        endpoint: `/internal/fleets/${encodeURIComponent(fleetId)}/cleanup/approve`,
+        env: context.env,
+        fetchImpl: context.fetchImpl,
+        method: "POST",
+      });
+    } catch (error) {
+      if (
+        error instanceof CliApiError &&
+        error.detail?.detail === "cleanup_plan_replan_required"
+      ) {
+        throw new Error(
+          `Fleet ${fleetId} has a legacy cleanup plan that the current runner cannot safely execute. Re-run the same fleet with --dry-run, review the generated resource policy, then approve the new plan hash.`,
+        );
+      }
+      throw error;
+    }
     context.stderr.write(`Cleaning ${approved.nodes.length} node(s)...\n`);
     const results = await mapLimit(approved.nodes, parallel, async (node) => {
       let cleanupResult;
@@ -2452,7 +2604,7 @@ async function fleetClean(args, context) {
         });
         const encodedPlan = Buffer.from(JSON.stringify(node.plan), "utf8").toString("base64");
         let result;
-        let duringCanaryChecks = 0;
+        let duringCanary = { failedChecks: 0, successfulChecks: 0 };
         try {
           result = await runFleetSshCapture({
             context,
@@ -2465,7 +2617,10 @@ async function fleetClean(args, context) {
             user: node.ssh_user,
           });
         } finally {
-          duringCanaryChecks = await stopCanary();
+          duringCanary = await stopCanary();
+        }
+        if (result.exitCode === 124) {
+          throw new Error(`cleanup_execution_deferred: ${result.stderr}`);
         }
         const payload = fleetCleanupJson(result.stdout);
         if (result.exitCode !== 0 || !payload.evidence) {
@@ -2485,7 +2640,8 @@ async function fleetClean(args, context) {
         }
         payload.evidence.management_ssh_canary = {
           before: true,
-          during_checks: duringCanaryChecks,
+          during_checks: duringCanary.successfulChecks,
+          during_failures: duringCanary.failedChecks,
           after: true,
         };
         cleanupResult = {
@@ -2498,7 +2654,11 @@ async function fleetClean(args, context) {
         cleanupResult = {
           cleanup_run_id: node.cleanup_run_id,
           succeeded: false,
-          ...(message.includes("cleanup_already_running") ? { deferred: true } : {}),
+          ...(/cleanup_already_running|cleanup_execution_deferred|SSH command timed out/i.test(
+            message,
+          )
+            ? { deferred: true }
+            : {}),
           error: message,
         };
       }
@@ -2543,12 +2703,16 @@ async function fleetClean(args, context) {
       env: context.env,
       fetchImpl: context.fetchImpl,
     });
-    if (fleetRecord.status !== "failed") {
-      throw new Error("Only a failed fleet can be replanned; resume an approved cleanup with --confirm-clean.");
+    if (!["failed", "cleaning"].includes(fleetRecord.status)) {
+      throw new Error("Only a failed or still-planning fleet can be replanned.");
     }
-    planningNodes = fleetRecord.nodes.filter((node) => node.status === "failed");
+    planningNodes = fleetRecord.nodes.filter((node) =>
+      fleetRecord.status === "failed"
+        ? node.status === "failed"
+        : ["pending_clean", "clean_planned"].includes(node.status),
+    );
     if (!planningNodes.length) {
-      throw new Error("This fleet has no failed cleanup members to replan.");
+      throw new Error("This fleet has no cleanup members eligible for replanning.");
     }
   } else {
     const ips = uniqueFleetIps(positionals);
@@ -2589,12 +2753,16 @@ async function fleetClean(args, context) {
           ...sourceUsers.map((user) => `--source-user ${shellQuote(user)}`),
           ...preserveUsers.map((user) => `--preserve-user ${shellQuote(user)}`),
         ].join(" ");
+        const nodePolicy = cleanupPolicyForNode(loadedPolicy, node, planningNodes.length);
+        const policyArgument = nodePolicy
+          ? ` --policy-b64 ${shellQuote(Buffer.from(JSON.stringify(nodePolicy), "utf8").toString("base64"))}`
+          : "";
         const result = await runFleetSshCapture({
           context,
           identityFile,
           ip: node.ip_address,
           port: sshPort,
-          remoteCommand: `sudo -n python3 - plan --management-ip ${node.ip_address} --management-user ${sshUser} --management-public-key-b64 ${managementPublicKeyB64}${accountArguments ? ` ${accountArguments}` : ""}`,
+          remoteCommand: `sudo -n python3 - plan --management-ip ${node.ip_address} --management-user ${sshUser} --management-public-key-b64 ${managementPublicKeyB64}${accountArguments ? ` ${accountArguments}` : ""}${policyArgument}`,
           stdin: runner.script,
           timeoutSeconds,
           user: sshUser,
@@ -2604,6 +2772,9 @@ async function fleetClean(args, context) {
           throw new Error(payload.error || result.stderr || "Cleanup plan failed.");
         }
         return {
+          ip: node.ip_address,
+          observations: payload.observations || {},
+          policyTemplate: payload.policy_template,
           plan: {
             fleet_node_id: node.id,
             ssh_user: sshUser,
@@ -2622,6 +2793,23 @@ async function fleetClean(args, context) {
       );
     }
     const plans = planned.map((result) => result.plan);
+    const policyTemplate = {
+      schema_version: 1,
+      nodes: { ...(loadedPolicy?.policy?.nodes || {}) },
+    };
+    const observations = {};
+    for (const result of [...planned].sort((left, right) => left.ip.localeCompare(right.ip))) {
+      policyTemplate.nodes[result.ip] = result.policyTemplate;
+      observations[result.ip] = result.observations;
+    }
+    const savedPolicyPath =
+      policyOut || join(getFleetConfigDir(context.env), `${fleetRecord.id}.cleanup-policy.json`);
+    await mkdir(dirname(savedPolicyPath), { recursive: true, mode: 0o700 });
+    await writeFile(savedPolicyPath, `${JSON.stringify(policyTemplate, null, 2)}\n`, {
+      encoding: "utf8",
+      mode: 0o600,
+    });
+    await chmod(savedPolicyPath, 0o600);
     recorded = await operatorRequest({
       body: { nodes: plans },
       endpoint: `/internal/fleets/${encodeURIComponent(fleetRecord.id)}/cleanup/plans`,
@@ -2629,10 +2817,13 @@ async function fleetClean(args, context) {
       fetchImpl: context.fetchImpl,
       method: "POST",
     });
+    recorded.policy_template = policyTemplate;
+    recorded.observations = observations;
+    recorded.policy_file = savedPolicyPath;
   } catch (error) {
     const planningError = fleetResultError(error);
     if (!createdFleet) {
-      throw new Error(`${planningError} Fleet ${fleetRecord.id} remains failed and can be replanned.`);
+      throw new Error(`${planningError} Fleet ${fleetRecord.id} can be replanned after correction.`);
     }
     let failedFleet;
     try {
@@ -2673,9 +2864,24 @@ async function fleetClean(args, context) {
       }
     }
     context.stdout.write(`Cleanup plan: ${recorded.plan_hash}\n`);
-    context.stdout.write(
-      `Approve exactly this plan with: ornn fleet clean ${fleetRecord.id} --identity-file ${shellQuote(identityFile)} --confirm-clean ${recorded.plan_hash}${sourceUsers.map((user) => ` --source-user ${shellQuote(user)}`).join("")}${preserveUsers.map((user) => ` --preserve-user ${shellQuote(user)}`).join("")}\n`
+    context.stdout.write(`Policy template: ${recorded.policy_file}\n`);
+    const unresolved = recorded.fleet.nodes.reduce(
+      (count, node) => count + (node.cleanup?.plan?.unresolved?.length || 0),
+      0,
+    );
+    const conflicts = recorded.fleet.nodes.reduce(
+      (count, node) => count + (node.cleanup?.plan?.conflicts?.length || 0),
+      0,
     );
+    if (unresolved || conflicts) {
+      context.stdout.write(
+        `Decisions required: ${unresolved}; policy conflicts: ${conflicts}. Edit the policy template and re-run the dry-run for fleet ${fleetRecord.id}.\n`,
+      );
+    } else {
+      context.stdout.write(
+        `Approve exactly this plan with: ornn fleet clean ${fleetRecord.id} --identity-file ${shellQuote(identityFile)} --confirm-clean ${recorded.plan_hash}\n`,
+      );
+    }
   }
   return 0;
 }
@@ -3431,12 +3637,11 @@ async function deriveFleetManagementPublicKey({ context, identityFile }) {
 }
 
 function startFleetSshCanary({ context, identityFile, ip, port, timeoutSeconds, user }) {
-  let failure = null;
+  let failedChecks = 0;
   let successfulChecks = 0;
   let inFlight = Promise.resolve();
   const check = () => {
     inFlight = inFlight.then(async () => {
-      if (failure) return;
       const exitCode = await runFleetSsh({
         context,
         identityFile,
@@ -3447,7 +3652,7 @@ function startFleetSshCanary({ context, identityFile, ip, port, timeoutSeconds,
         user,
       });
       if (exitCode !== 0) {
-        failure = new Error("Management SSH canary failed during cleanup.");
+        failedChecks += 1;
       } else {
         successfulChecks += 1;
       }
@@ -3459,8 +3664,7 @@ function startFleetSshCanary({ context, identityFile, ip, port, timeoutSeconds,
   return async () => {
     clearInterval(timer);
     await inFlight;
-    if (failure) throw failure;
-    return successfulChecks;
+    return { failedChecks, successfulChecks };
   };
 }
 
@@ -5940,14 +6144,17 @@ function inventoryAvailableGpuCount(record) {
 
 function normalizeInventoryListing(record) {
   const gpuCount = inventoryAvailableGpuCount(record);
+  const facility = record.site_nickname ?? record.facility ?? record.site ?? "Unknown facility";
+  // Customer availability never exposes real supplier/operator identity.
+  const operator = "Ornn Compute";
   return {
     id: record.id,
     kind: "inventory",
     source: "primary",
     gpu_type: record.gpu_type ?? "GPU",
     gpu_count: gpuCount || null,
-    operator: record.site_operator ?? record.operator ?? "Unknown operator",
-    facility: record.site_nickname ?? record.facility ?? record.site ?? "Unknown facility",
+    operator,
+    facility,
     location: record.location ?? record.region ?? null,
     network: record.fabric_type ?? record.network_hardware ?? record.internet ?? null,
     start_date: record.available_from ?? null,
@@ -5955,7 +6162,8 @@ function normalizeInventoryListing(record) {
     price_per_gpu_hour: resolveBuyNowUsdPerGpuHour(record),
     checkout_url: `/checkout?inventory=${encodeURIComponent(record.id)}`,
     marketplace_url: `/marketplace/${encodeURIComponent(record.id)}`,
-    inventory: record,
+    // Strip supplier identity before echoing the raw inventory row in --json.
+    inventory: { ...record, site_operator: operator },
   };
 }