@celilo/cli 0.3.25 → 0.3.26

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@celilo/cli",
-  "version": "0.3.25",
+  "version": "0.3.26",
   "description": "Celilo — home lab orchestration CLI",
   "type": "module",
   "bin": {

package/src/cli/commands/system-update.ts

@@ -159,9 +159,13 @@ async function buildSnapshots(
  * - `health` calls `runModuleHealthCheck`.
  * - `snapshotCeliloDb` calls `createSystemStateBackup`.
  */
-function buildOps(registry: RegistryClient): OrchestratorOps {
+function buildOps(registry: RegistryClient, wasDeployed: Set<string>): OrchestratorOps {
   return {
     backup: async (moduleId, _updateId) => {
+      // IMPORTED modules don't have running state to back up; skip
+      // silently. The orchestrator already gates on `!noBackup`, so
+      // this is the second layer (per-module rather than system-wide).
+      if (!wasDeployed.has(moduleId)) return { ok: true };
       const db = getDb();
       const mod = db.select().from(modules).where(eq(modules.id, moduleId)).get();
       if (!mod) return { ok: false, error: `module ${moduleId} not found` };
@@ -202,6 +206,12 @@ function buildOps(registry: RegistryClient): OrchestratorOps {
       }
     },
     deploy: async (id) => {
+      // IMPORTED modules weren't deployed before this run; we just
+      // refreshed their source files via `upgrade`. Don't try to
+      // deploy them here — that would surprise the operator who
+      // hasn't asked for a deploy. They'll get a `module deploy <id>`
+      // todo finding from the audit on the next run.
+      if (!wasDeployed.has(id)) return { ok: true };
       const db = getDb();
       // The deploy interview runs through the bus; if config is
       // missing the deploy hangs waiting for a responder. system-update
@@ -211,6 +221,19 @@ function buildOps(registry: RegistryClient): OrchestratorOps {
       return result.success ? { ok: true } : { ok: false, error: result.error };
     },
     health: async (id) => {
+      // Same gating as deploy: there's no live deployment to health-
+      // check for an IMPORTED module that we just refreshed. Return
+      // pass with a "not deployed" detail so the orchestrator's
+      // module-step record doesn't say "health: skipped" (which would
+      // imply a check ran and skipped its assertions).
+      if (!wasDeployed.has(id)) {
+        // The orchestrator's HealthStatus type uses 'healthy' /
+        // 'degraded' / 'unhealthy' / 'error' / 'no-checks'. There's
+        // no "skipped" — closest meaningful value for "we didn't
+        // run the check on purpose" is 'no-checks' (semantically:
+        // "the module didn't have any health checks to run").
+        return { status: 'no-checks', detail: 'not deployed; health check skipped' };
+      }
       // We can call the existing health-runner directly today — that's
       // already a clean service.
       const db = getDb();
@@ -249,17 +272,20 @@ function formatResult(result: SystemUpdateResult): string {
   );
   lines.push(`  backups: ${result.backupsCreated ? 'created' : 'skipped'}`);
 
-  // Surface audit findings inline. DRIFT means "something is drifting
-  // but it didn't block the update" — without listing the findings the
-  // operator has no idea what; they'd have to run `celilo system audit`
-  // as a follow-up. Show them here so it's a single round-trip.
-  // BLOCKED also gets the listing (the orchestrator already short-
-  // circuited to skip the run, but the operator still needs to know
-  // why — formatResult is the friendly path for that too).
+  // Surface audit findings inline. Three severity tiers:
+  //   blocked → ✗   gates the run (orchestrator short-circuits)
+  //   drift   → ▸   informational; the system moved away from a
+  //                 desired state, but the run still proceeded
+  //   todo    → ➤   next-step reminders (e.g. "you imported X but
+  //                 haven't deployed it yet"); never escalates the
+  //                 verdict.
+  // Each tier renders separately so operators can scan for what
+  // needs attention vs. what's just a friendly nudge.
   if (result.audit.findings.length > 0) {
     lines.push('');
-    const drift = result.audit.findings.filter((f) => f.severity === 'drift');
     const blocked = result.audit.findings.filter((f) => f.severity === 'blocked');
+    const drift = result.audit.findings.filter((f) => f.severity === 'drift');
+    const todo = result.audit.findings.filter((f) => f.severity === 'todo');
     if (blocked.length > 0) {
       lines.push(`  BLOCKED (${blocked.length}):`);
       for (const f of blocked) {
@@ -274,6 +300,13 @@ function formatResult(result: SystemUpdateResult): string {
         if (f.remediation) lines.push(`        → ${f.remediation}`);
       }
     }
+    if (todo.length > 0) {
+      lines.push(`  Todos (${todo.length}, next-step reminders):`);
+      for (const f of todo) {
+        lines.push(`    ➤ [${f.category}] ${f.subject}: ${f.message}`);
+        if (f.remediation) lines.push(`        → ${f.remediation}`);
+      }
+    }
   }
 
   if (result.modules.length > 0) {
@@ -303,12 +336,28 @@ export async function handleSystemUpdate(
 
   const db = getDb();
   const installed = db.select().from(modules).all();
-  const deployedModules = installed.filter((m) => ['INSTALLED', 'VERIFIED'].includes(m.state));
+  // Include IMPORTED modules in the upgrade scope so a fresh
+  // `module import <name>` followed (later, possibly weeks later) by
+  // `system update` actually picks up registry-side updates for
+  // not-yet-deployed modules. The deploy and health steps are gated
+  // on prior state in buildOps so we don't deploy something the
+  // operator hasn't asked us to deploy.
+  // INSTALLED + VERIFIED + IMPORTED is the upgrade-eligible set;
+  // ERROR / DEPLOYING / GENERATING / UNINSTALLING are skipped (the
+  // operator needs to handle those manually first).
+  const upgradeEligibleStates = new Set(['INSTALLED', 'VERIFIED', 'IMPORTED']);
+  const upgradableModules = installed.filter((m) => upgradeEligibleStates.has(m.state));
+  // Track which modules were already deployed BEFORE this run.
+  // Anything else gets the upgrade only — no deploy / health / backup
+  // attempts (those would be no-ops at best, surprising at worst).
+  const wasDeployed = new Set(
+    installed.filter((m) => ['INSTALLED', 'VERIFIED'].includes(m.state)).map((m) => m.id),
+  );
   const registry = new RegistryClient();
 
   // Build the dep graph from installed manifests.
-  const graph = buildModuleGraph(deployedModules.map((m) => m.manifestData as ModuleManifest));
-  const snapshots = await buildSnapshots(db, deployedModules, registry);
+  const graph = buildModuleGraph(upgradableModules.map((m) => m.manifestData as ModuleManifest));
+  const snapshots = await buildSnapshots(db, upgradableModules, registry);
 
   // Build audit deps. (Mostly mirrors system-audit.ts; could be refactored
   // into a shared helper in Phase 5.)
@@ -351,13 +400,13 @@ export async function handleSystemUpdate(
       db,
     },
     capabilityAbi: {
-      modules: deployedModules.map((m) => ({
+      modules: upgradableModules.map((m) => ({
         id: m.id,
         manifest: m.manifestData as ModuleManifest,
       })),
     },
     terraformPlan: {
-      modules: deployedModules.map((m) => ({
+      modules: upgradableModules.map((m) => ({
         id: m.id,
         terraformDir: existsSync(join(m.sourcePath, 'generated', 'terraform'))
           ? join(m.sourcePath, 'generated', 'terraform')
@@ -366,7 +415,7 @@ export async function handleSystemUpdate(
       run: async () => ({ exitCode: 0, stdout: '', stderr: '' }),
     },
     moduleVersions: {
-      installed: deployedModules.map((m) => ({ id: m.id, version: m.version })),
+      installed: upgradableModules.map((m) => ({ id: m.id, version: m.version })),
       fetcher: async (id: string) => {
         const entries = await registry.getIndex(id);
         const top = registry.latestVersion(entries);
@@ -374,7 +423,7 @@ export async function handleSystemUpdate(
       },
     },
     moduleConfigs: {
-      modules: deployedModules.map((m) => ({
+      modules: upgradableModules.map((m) => ({
         id: m.id,
         manifest: m.manifestData as ModuleManifest,
         configs: configsByModule.get(m.id) ?? {},
@@ -382,7 +431,7 @@ export async function handleSystemUpdate(
     },
     health: { results: healthResults },
     backups: {
-      modules: deployedModules.map((m) => ({
+      modules: upgradableModules.map((m) => ({
         id: m.id,
         manifest: m.manifestData as ModuleManifest,
         lastSuccessfulBackupAt: latestBackupByModule.get(m.id) ?? null,
@@ -415,7 +464,7 @@ export async function handleSystemUpdate(
       updateId: crypto.randomUUID(),
       audit,
       willSelfUpdate: false, // computed at run time; the dry-run preview is best-effort
-      modules: deployedModules
+      modules: upgradableModules
         .filter((m) => {
           const snap = snapshots.get(m.id);
           return snap?.latestVersion && snap.latestVersion !== snap.installedVersion;
@@ -482,7 +531,7 @@ Then re-run system update.`,
     }
   }
 
-  const ops = buildOps(registry);
+  const ops = buildOps(registry, wasDeployed);
 
   const result = await runSystemUpdate({
     audit: auditDeps,

package/src/cli/tui/icons.ts

@@ -9,6 +9,7 @@ export interface SeverityVisual {
 export const SEVERITY_VISUALS: Record<DriftSeverity, SeverityVisual> = {
   blocked: { icon: '×', color: 'red', label: 'BLOCKED' },
   drift: { icon: '▲', color: 'yellow', label: 'DRIFT' },
+  todo: { icon: '➤', color: 'gray', label: 'TODO' },
 };
 
 export const VERDICT_VISUALS: Record<AuditVerdict, SeverityVisual> = {
@@ -17,6 +18,11 @@ export const VERDICT_VISUALS: Record<AuditVerdict, SeverityVisual> = {
   READY: { icon: '✓', color: 'green', label: 'READY' },
 };
 
+// Lower rank = higher priority (sorts first when listing). todo is
+// the lowest priority — operators can scan past them when triaging
+// what actually needs attention.
 export function severityRank(s: DriftSeverity): number {
-  return s === 'blocked' ? 0 : 1;
+  if (s === 'blocked') return 0;
+  if (s === 'drift') return 1;
+  return 2; // todo
 }

package/src/services/audit/types.test.ts

@@ -17,6 +17,14 @@ const blocked: DriftFinding = {
   subject: 'lunacycle',
 };
 
+const todo: DriftFinding = {
+  category: 'undeployed_modules',
+  severity: 'todo',
+  code: 'module_undeployed',
+  message: 'namecheap: imported but not deployed (state: IMPORTED)',
+  subject: 'namecheap',
+};
+
 describe('computeVerdict', () => {
   test('READY when no findings', () => {
     expect(computeVerdict([])).toBe('READY');
@@ -33,4 +41,22 @@ describe('computeVerdict', () => {
   test('BLOCKED takes precedence regardless of order', () => {
     expect(computeVerdict([blocked, drift])).toBe('BLOCKED');
   });
+
+  // Pinning down the new severity tier: todo findings are
+  // informational reminders and never escalate the verdict beyond
+  // READY. The whole point of this severity level is to let
+  // categories like undeployed_modules and unconfigured_modules
+  // surface a TODO list without making `system update` look like
+  // it has unfinished work.
+  test('READY when only todo-severity findings', () => {
+    expect(computeVerdict([todo, { ...todo, subject: 'caddy' }])).toBe('READY');
+  });
+
+  test('DRIFT when both drift and todo findings exist (drift wins)', () => {
+    expect(computeVerdict([todo, drift])).toBe('DRIFT');
+  });
+
+  test('BLOCKED still wins over todos', () => {
+    expect(computeVerdict([todo, drift, blocked])).toBe('BLOCKED');
+  });
 });

package/src/services/audit/types.ts

@@ -3,14 +3,22 @@
  *
  * `runAudit` returns a `SystemAuditReport` aggregating findings from
  * each drift category. Each finding has a category (one of
- * `DriftCategory`), a severity (`drift` or `blocked`), a stable
+ * `DriftCategory`), a severity (`todo`, `drift`, or `blocked`), a stable
  * machine-readable code, and a human-readable message + suggested
  * remediation.
  *
  * The overall verdict is computed from the findings:
  * - any `blocked` finding → BLOCKED
  * - any `drift` finding (no blocked) → DRIFT
- * - no findings → READY
+ * - only `todo` findings (or none) → READY
+ *
+ * `todo` exists because some categories surface "next-step reminders"
+ * rather than actual divergence — `unconfigured_modules` and
+ * `undeployed_modules` are the canonical examples. The operator
+ * imported a module and hasn't gotten around to configuring/deploying;
+ * that's a TODO list, not a drifted system. Calling it DRIFT
+ * overloaded the term and made `system update` look like it had
+ * unfinished work even when it didn't.
  */
 
 export type DriftCategory =
@@ -29,7 +37,7 @@ export type DriftCategory =
   | 'services_reachable'
   | 'machines_reachable';
 
-export type DriftSeverity = 'drift' | 'blocked';
+export type DriftSeverity = 'todo' | 'drift' | 'blocked';
 
 export type AuditVerdict = 'READY' | 'DRIFT' | 'BLOCKED';
 
@@ -81,10 +89,12 @@ export interface SystemAuditReport {
 }
 
 /**
- * Compute the overall verdict from a set of findings.
+ * Compute the overall verdict from a set of findings. `todo` findings
+ * never escalate beyond READY — they're informational reminders, not
+ * a signal that the system has moved away from a desired state.
  */
 export function computeVerdict(findings: DriftFinding[]): AuditVerdict {
   if (findings.some((f) => f.severity === 'blocked')) return 'BLOCKED';
-  if (findings.length > 0) return 'DRIFT';
+  if (findings.some((f) => f.severity === 'drift')) return 'DRIFT';
   return 'READY';
 }

package/src/services/audit/unconfigured-modules.test.ts

@@ -9,7 +9,7 @@ describe('auditUnconfiguredModules', () => {
     expect(result).toHaveLength(1);
     expect(result[0]).toMatchObject({
       category: 'unconfigured_modules',
-      severity: 'drift',
+      severity: 'todo',
       code: 'module_unconfigured',
       actionable: false,
       subject: 'authentik',

package/src/services/audit/unconfigured-modules.ts

@@ -41,9 +41,13 @@ export async function auditUnconfiguredModules(
     if (ALREADY_DEPLOYED.has(m.state)) continue;
     if (m.configCount > 0) continue;
 
+    // todo: same reasoning as undeployed_modules — never running the
+    // configuration interview is a next-step reminder, not divergence
+    // from a desired state. Won't escalate the audit verdict beyond
+    // READY.
     findings.push({
       category: 'unconfigured_modules',
-      severity: 'drift',
+      severity: 'todo',
       code: 'module_unconfigured',
       message: `${m.id}: never configured (no config rows in DB)`,
       details: [

package/src/services/audit/undeployed-modules.test.ts

@@ -19,7 +19,7 @@ describe('auditUndeployedModules', () => {
     expect(result).toHaveLength(1);
     expect(result[0]).toMatchObject({
       category: 'undeployed_modules',
-      severity: 'drift',
+      severity: 'todo',
       code: 'module_undeployed',
       remediation: 'celilo module deploy authentik',
       actionable: true,

package/src/services/audit/undeployed-modules.ts

@@ -57,9 +57,15 @@ export async function auditUndeployedModules(
       continue;
     }
 
+    // todo: this is a next-step reminder, not a divergence from a
+    // desired state. The operator imported the module and hasn't
+    // deployed yet — that's a TODO list, not "the system has drifted".
+    // ERROR-state modules (handled above) DO stay severity=blocked
+    // because the operator's prior intent was to deploy, the deploy
+    // failed, and the system is now stuck pending intervention.
     findings.push({
       category: 'undeployed_modules',
-      severity: 'drift',
+      severity: 'todo',
       code: 'module_undeployed',
       message: `${m.id}: imported but not deployed (state: ${m.state})`,
       remediation: `celilo module deploy ${m.id}`,