@ornncompute/cli 0.1.7 → 0.1.9

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 <non-root-linux-user>] [--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 +228,73 @@ 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`, or the
+  explicit non-root account supplied with `--ssh-user`. It 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.
+  Coverage gates also inventory listeners, GPU and container-runtime visibility,
+  non-block mounts, advanced SSH sources, filesystem scope, root-owned workloads,
+  network state, groups/sudo, IPC, swap/RAM, and active block stacks. A verified
+  gate is kept automatically. A present or uninspectable gate must be explicitly
+  ignored with a reason, so the receipt exposes the excluded scope.
+- `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.
+  Kept files and directories are content-hashed; process identity, container
+  replacement state, and exact drive topology are rechecked. 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. Logical clean is offered only when
+  every filesystem in the approved topology is mounted and can be emptied and
+  verified; it never substitutes `wipefs` for data deletion. Discard and NVMe
+  sanitize require device-reported support, and strong actions refuse active or
+  unverified block stacks. `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-v7 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,
+  live-revalidates each receipt under the current enrollment attempt, then
+  issues and revokes one cleanup- and attempt-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.
+  in Compute. A private executor credential prevents two CLIs from sharing an
+  active lease. Retries reconcile nodes already registered by the same attempt
+  before rerunning cleanup or installation. 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
@@ -266,8 +312,10 @@ Fleet handoff is an ordered, Compute-backed workflow:
   node; Commerce is stamped only after the entire fleet reports deployed. A
   Commerce reservation can be linked to only one fleet.
 
-Compute Postgres is authoritative; the mode-0600 files under
-`~/.config/ornn/fleets/` are only a secret-free cache. Cleanup proves “no
+Compute Postgres is authoritative. The mode-0600 files under
+`~/.config/ornn/fleets/` include a short-lived enrollment resume credential;
+treat them as sensitive and do not copy them into tickets or source control.
+Cleanup proves “no
 observable residue under the verification scope,” not certified physical-media
 erasure. Root is never the preserved management account. `--confirm-takeover`
 names only fleet IPs whose existing node claims may be replaced. The IB-island

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ornncompute/cli",
-  "version": "0.1.7",
+  "version": "0.1.9",
   "description": "Command-line interface for Ornn compute access workflows.",
   "type": "module",
   "bin": {

package/src/api-client.mjs

@@ -107,7 +107,7 @@ export function resolveComputeBaseUrl({ env = process.env } = {}) {
   const base = env.ORNN_COMPUTE_BASE_URL?.trim();
   if (!base) {
     throw new CliApiError(
-      "Set ORNN_COMPUTE_BASE_URL to the internal compute service origin to run operator commands.",
+      "Set ORNN_COMPUTE_BASE_URL to the internal compute service origin to run operator commands."
     );
   }
   return requireValidBaseUrl(base, "ORNN_COMPUTE_BASE_URL");
@@ -173,7 +173,7 @@ export function resolveInternalReviewSecret({ env = process.env } = {}) {
   const secret = env.ORNN_INTERNAL_REVIEW_SECRET?.trim();
   if (!secret) {
     throw new CliApiError(
-      "Set ORNN_INTERNAL_REVIEW_SECRET to authenticate operator commands, or run `ornn login` as Ornn staff.",
+      "Set ORNN_INTERNAL_REVIEW_SECRET to authenticate operator commands, or run `ornn login` as Ornn staff."
     );
   }
   return secret;
@@ -267,7 +267,7 @@ export async function operatorRequest({
   throw new CliApiError(
     "Operator commands need Ornn staff auth: run `ornn login` as an internal user " +
       "(ORNN_AUTH_BASE_URL), or set ORNN_INTERNAL_REVIEW_SECRET (+ ORNN_COMPUTE_BASE_URL) for direct compute access.",
-    { status: 401 },
+    { status: 401 }
   );
 }
 
@@ -287,7 +287,7 @@ export async function webOperatorRequest({
   if (!session?.accessToken) {
     throw new CliApiError(
       "Commerce-backed deployment requires Ornn staff auth through Web. Run `ornn login` first.",
-      { status: 401 },
+      { status: 401 }
     );
   }
   return operatorRequestViaStaffSession({ body, endpoint, env, fetchImpl, method });
@@ -313,7 +313,7 @@ function requireValidBaseUrl(raw, sourceName) {
   const stripped = raw.replace(/\/+$/, "");
   if (!stripped) {
     throw new CliApiError(
-      `${sourceName} must be a valid origin (got empty value after removing trailing slashes).`,
+      `${sourceName} must be a valid origin (got empty value after removing trailing slashes).`
     );
   }
   let parsed;
@@ -325,7 +325,7 @@ function requireValidBaseUrl(raw, sourceName) {
   // Path/search/hash are dropped when composing absolute endpoints against this base.
   if (parsed.pathname !== "/" || parsed.search || parsed.hash) {
     throw new CliApiError(
-      `${sourceName} must be a valid origin with no path (got ${JSON.stringify(raw)}).`,
+      `${sourceName} must be a valid origin with no path (got ${JSON.stringify(raw)}).`
     );
   }
   return stripped;
@@ -337,9 +337,7 @@ function parseJson(text, url, response) {
   } catch {
     const contentType = response?.headers?.get?.("content-type") || "";
     const isHtml = /html/i.test(contentType) || /^\s*</.test(text);
-    const hint = isHtml
-      ? " The Ornn host may not have this CLI endpoint deployed yet."
-      : "";
+    const hint = isHtml ? " The Ornn host may not have this CLI endpoint deployed yet." : "";
     throw new CliApiError(`Ornn returned a non-JSON response from ${url}.${hint}`, {
       status: response?.status,
     });
@@ -351,26 +349,40 @@ async function handleAuthExpiry({ env, response, hadToken }) {
     return;
   }
   await clearAuthSession({ env });
-  throw new CliApiError("Your Ornn session has expired or was revoked. Run `ornn login` to sign in again.", {
-    status: 401,
-  });
+  throw new CliApiError(
+    "Your Ornn session has expired or was revoked. Run `ornn login` to sign in again.",
+    {
+      status: 401,
+    }
+  );
 }
 
 function detailMessage(detail) {
   if (typeof detail === "string") {
     return detail;
   }
+  if (Array.isArray(detail)) {
+    return detail
+      .map((item) => detailMessage(item))
+      .filter(Boolean)
+      .join("; ");
+  }
   if (!detail || typeof detail !== "object") {
     return "";
   }
-  if (typeof detail.message === "string") {
-    return detail.message;
-  }
-  if (typeof detail.summary === "string") {
-    return detail.summary;
-  }
-  if (typeof detail.code === "string") {
-    return detail.code;
+  const message = [detail.msg, detail.message, detail.summary, detail.code].find(
+    (value) => typeof value === "string" && value.trim()
+  );
+  const location = Array.isArray(detail.loc)
+    ? detail.loc
+        .filter((value) => typeof value === "string" || typeof value === "number")
+        .map(String)
+        .join(".")
+    : typeof detail.loc === "string"
+      ? detail.loc.trim()
+      : "";
+  if (message) {
+    return location ? `${location}: ${message.trim()}` : message.trim();
   }
-  return "";
+  return detailMessage(detail.detail ?? detail.errors);
 }