@celilo/cli 0.3.27 → 0.3.29

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@celilo/cli",
-  "version": "0.3.27",
+  "version": "0.3.29",
   "description": "Celilo — home lab orchestration CLI",
   "type": "module",
   "bin": {

package/src/cli/commands/storage-add-local.test.ts

@@ -0,0 +1,66 @@
+/**
+ * Tests for `storage add local`'s pre-save write probe.
+ *
+ * Background: an operator on celilo-mgmt typed `/var/backups/celilo`
+ * for the path. The CLI accepted it, saved a storage row, then the
+ * heavyweight verify step failed with EACCES — leaving an unverified
+ * row that confused subsequent `system update` runs. The probe
+ * catches unwriteable paths before any persistent state is created.
+ */
+
+import { afterEach, beforeEach, describe, expect, test } from 'bun:test';
+import { mkdirSync, mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { probePathWriteable } from './storage-add-local';
+
+describe('probePathWriteable', () => {
+  let tempRoot: string;
+
+  beforeEach(() => {
+    tempRoot = mkdtempSync(join(tmpdir(), 'celilo-probe-test-'));
+  });
+
+  afterEach(() => {
+    rmSync(tempRoot, { recursive: true, force: true });
+  });
+
+  test('returns null for a writeable existing directory', () => {
+    expect(probePathWriteable(tempRoot)).toBeNull();
+  });
+
+  test("creates parent dirs that don't exist yet (mkdir -p semantics)", () => {
+    // The probe's mkdirSync uses `recursive: true`, so a path several
+    // levels below the temp root works on the first try. This is the
+    // common case for the user's typed path under their data dir.
+    const deep = join(tempRoot, 'a', 'b', 'c', 'backups');
+    expect(probePathWriteable(deep)).toBeNull();
+  });
+
+  test('returns an error message for a path under a read-only ancestor', () => {
+    const readOnly = join(tempRoot, 'readonly');
+    mkdirSync(readOnly);
+    // 0o500 = read+execute for owner, no write. mkdirSync into it
+    // should EACCES.
+    require('node:fs').chmodSync(readOnly, 0o500);
+    try {
+      const target = join(readOnly, 'celilo-backups');
+      const err = probePathWriteable(target);
+      expect(err).not.toBeNull();
+      expect(err).toContain('EACCES');
+    } finally {
+      // Restore permissions so afterEach can clean up.
+      require('node:fs').chmodSync(readOnly, 0o700);
+    }
+  });
+
+  test('cleans up the probe directory on success (no leftover state)', () => {
+    const dir = join(tempRoot, 'check');
+    expect(probePathWriteable(dir)).toBeNull();
+    // The probe creates `<dir>/.celilo-write-probe` then removes it.
+    // The directory itself stays (mkdir -p), but the probe sentinel
+    // doesn't.
+    const { existsSync } = require('node:fs');
+    expect(existsSync(join(dir, '.celilo-write-probe'))).toBe(false);
+  });
+});

package/src/cli/commands/storage-add-local.ts

@@ -3,6 +3,7 @@
  * Configure a local filesystem backup storage destination
  */
 
+import { mkdirSync, rmSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { join, resolve } from 'node:path';
 import { getDataDir } from '../../config/paths';
@@ -15,6 +16,27 @@ import { celiloIntro, celiloOutro, promptConfirm, promptText } from '../prompts'
 import type { CommandResult } from '../types';
 import { validateRequired } from '../validators';
 
+/**
+ * Best-effort write probe. Attempts to mkdir -p a `.celilo-write-probe`
+ * subdir, then deletes it. Returns null on success, or a one-line
+ * error message on failure (suitable for surfacing to the operator).
+ *
+ * Exists so we can catch unwriteable paths AT INTERVIEW TIME instead
+ * of saving a storage row, failing the heavyweight verify step, and
+ * leaving a dangling unverified row that confuses subsequent
+ * `system update` runs (the regression that prompted this work).
+ */
+export function probePathWriteable(path: string): string | null {
+  const probe = join(path, '.celilo-write-probe');
+  try {
+    mkdirSync(probe, { recursive: true });
+    rmSync(probe, { recursive: true, force: true });
+    return null;
+  } catch (err) {
+    return err instanceof Error ? err.message : String(err);
+  }
+}
+
 /**
  * Expand leading tilde to the user's home directory.
  * Node.js fs functions don't expand ~ like the shell does.
@@ -55,16 +77,46 @@ export async function handleStorageAddLocal(
     // looked authoritative but required root to write.
     const defaultPath = join(getDataDir(), 'backups');
 
-    const path =
-      flagPath ??
-      (await promptText({
-        message: 'Storage directory path:',
-        defaultValue: defaultPath,
-        placeholder: defaultPath,
-        validate: validateRequired('Storage path'),
-      }));
-
-    const resolvedPath = resolve(expandTilde(path));
+    // Probe writability BEFORE saving the storage row. The previous
+    // flow (save → verify → fail with EACCES → leave dangling
+    // unverified row) was confusing and required the operator to know
+    // about `storage verify` to recover. Probing here means an
+    // unwriteable path never produces persistent state.
+    //
+    // Non-interactive (--name + --path): probe once, fail loud.
+    // Interactive: re-prompt until a writeable path is supplied.
+    let resolvedPath: string;
+    if (flagPath !== undefined) {
+      resolvedPath = resolve(expandTilde(flagPath));
+      const writeError = probePathWriteable(resolvedPath);
+      if (writeError !== null) {
+        return {
+          success: false,
+          error: `Path '${resolvedPath}' is not writeable: ${writeError}\n\nTry a path under your home directory (e.g. '${defaultPath}') or run with elevated permissions.`,
+        };
+      }
+    } else {
+      let candidate: string | undefined;
+      while (candidate === undefined) {
+        const typed = await promptText({
+          message: 'Storage directory path:',
+          defaultValue: defaultPath,
+          placeholder: defaultPath,
+          validate: validateRequired('Storage path'),
+        });
+        const resolved = resolve(expandTilde(typed));
+        const writeError = probePathWriteable(resolved);
+        if (writeError === null) {
+          candidate = resolved;
+          break;
+        }
+        console.log(
+          `\n✗ Path '${resolved}' is not writeable: ${writeError}\nTry a path under your home directory (default '${defaultPath}' works without sudo).\n`,
+        );
+        // Loop continues — re-prompt with the same default suggestion.
+      }
+      resolvedPath = candidate;
+    }
 
     const storage = await addBackupStorage({
       name,
@@ -78,6 +130,10 @@ export async function handleStorageAddLocal(
     const { result } = await verifyBackupStorage(storage.id);
 
     if (!result.success) {
+      // Should be unreachable in interactive mode (the pre-save probe
+      // covers the EACCES path); could still fire for less-common
+      // verify failures (full disk, etc.). Keep the dangling-row
+      // recovery hint as a safety net.
       console.log(`\n✗ Verification failed: ${result.message}`);
       celiloOutro(
         `Storage '${storage.storageId}' added but not verified.\n\nFix the path and re-verify: celilo storage verify ${storage.storageId}`,

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

@@ -189,12 +189,14 @@ async function buildAuditDeps(onProgress?: (msg: string) => void) {
 
   const installedConfigs = deployedModules.map((m) => ({
     id: m.id,
+    state: m.state,
     manifest: m.manifestData as ModuleManifest,
     configs: configsByModule.get(m.id) ?? {},
   }));
 
   const installedBackupInfo = deployedModules.map((m) => ({
     id: m.id,
+    state: m.state,
     manifest: m.manifestData as ModuleManifest,
     lastSuccessfulBackupAt: latestBackupByModule.get(m.id) ?? null,
   }));
@@ -402,6 +404,7 @@ async function buildAuditDeps(onProgress?: (msg: string) => void) {
     capabilityAbi: {
       modules: deployedModules.map((m) => ({
         id: m.id,
+        state: m.state,
         manifest: m.manifestData as ModuleManifest,
       })),
     },

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

@@ -12,7 +12,12 @@
  */
 
 import { describe, expect, test } from 'bun:test';
-import { type BackupStorageLike, checkBackupStoragePreflight } from './system-update';
+import type { ModuleSnapshot } from '../../services/update/orchestrator';
+import {
+  type BackupStorageLike,
+  checkBackupStoragePreflight,
+  shouldTakeBackup,
+} from './system-update';
 
 const verified = (id: string): BackupStorageLike => ({ storageId: id, verified: true });
 const unverified = (id: string): BackupStorageLike => ({ storageId: id, verified: false });
@@ -101,3 +106,83 @@ describe('checkBackupStoragePreflight', () => {
     expect(result.message).toContain('Verified: home-backups, s3-cold');
   });
 });
+
+const snap = (
+  id: string,
+  installedVersion: string,
+  latestVersion: string | null,
+): ModuleSnapshot => ({
+  id,
+  installedVersion,
+  latestVersion,
+  installedProvides: {},
+  pendingRequires: {},
+});
+
+const snapshotsOf = (...entries: ModuleSnapshot[]): Map<string, ModuleSnapshot> =>
+  new Map(entries.map((s) => [s.id, s]));
+
+describe('shouldTakeBackup', () => {
+  test('false when no module updates at all', () => {
+    expect(
+      shouldTakeBackup({
+        snapshots: snapshotsOf(snap('caddy', '2.0.0+5', '2.0.0+5')),
+        wasDeployed: new Set(['caddy']),
+      }),
+    ).toBe(false);
+  });
+
+  test('true when a deployed module has a registry-newer version', () => {
+    expect(
+      shouldTakeBackup({
+        snapshots: snapshotsOf(snap('caddy', '2.0.0+5', '2.0.0+6')),
+        wasDeployed: new Set(['caddy']),
+      }),
+    ).toBe(true);
+  });
+
+  // The exact case the operator hit on celilo-mgmt: namecheap was
+  // imported but never deployed; system update wanted to back up the
+  // celilo DB even though no live state would be touched.
+  test('false when only IMPORTED modules have updates (live state untouched)', () => {
+    expect(
+      shouldTakeBackup({
+        snapshots: snapshotsOf(snap('namecheap', '3.1.0+10', '3.1.1+4')),
+        wasDeployed: new Set(), // nothing deployed
+      }),
+    ).toBe(false);
+  });
+
+  test('true when any deployed module needs updating, regardless of imported ones', () => {
+    expect(
+      shouldTakeBackup({
+        snapshots: snapshotsOf(
+          snap('namecheap', '3.1.0+10', '3.1.1+4'), // imported, has update
+          snap('caddy', '2.0.0+5', '2.0.0+6'), // deployed, has update
+        ),
+        wasDeployed: new Set(['caddy']),
+      }),
+    ).toBe(true);
+  });
+
+  test('false when registry has no info (latestVersion null)', () => {
+    // Network failure — we don't know if updates exist. Default to
+    // "no backup needed" rather than blocking the run; the per-module
+    // upgrade step will report individual failures if any.
+    expect(
+      shouldTakeBackup({
+        snapshots: snapshotsOf(snap('caddy', '2.0.0+5', null)),
+        wasDeployed: new Set(['caddy']),
+      }),
+    ).toBe(false);
+  });
+
+  test('false on empty snapshots', () => {
+    expect(
+      shouldTakeBackup({
+        snapshots: new Map(),
+        wasDeployed: new Set(),
+      }),
+    ).toBe(false);
+  });
+});

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

@@ -148,6 +148,29 @@ Or skip the safety net entirely (the CLI self-update still runs):
   };
 }
 
+/**
+ * Should `system update` take a celilo-DB snapshot for this run?
+ * True iff at least one module that's currently DEPLOYED has a
+ * registry-newer version waiting. Pure data in / data out.
+ *
+ * Excludes IMPORTED-but-not-yet-deployed modules — refreshing their
+ * on-disk source has no impact on live state, so the safety net
+ * isn't needed (and an operator who hasn't configured storage yet
+ * shouldn't be blocked by a snapshot they don't need).
+ */
+export function shouldTakeBackup(input: {
+  snapshots: Map<string, ModuleSnapshot>;
+  wasDeployed: Set<string>;
+}): boolean {
+  for (const [id, s] of input.snapshots) {
+    if (!input.wasDeployed.has(id)) continue;
+    if (!s.latestVersion) continue;
+    if (s.latestVersion === s.installedVersion) continue;
+    return true;
+  }
+  return false;
+}
+
 function readInstalledCliVersion(): string {
   const here = dirname(fileURLToPath(import.meta.url));
   const candidates = [
@@ -363,7 +386,12 @@ function buildOps(registry: RegistryClient, wasDeployed: Set<string>): Orchestra
 function formatResult(result: SystemUpdateResult): string {
   const lines: string[] = [];
   lines.push('');
-  lines.push(`update ${result.updateId} → ${result.ok ? 'OK' : 'FAILED'}`);
+  // The updateId is for audit / journal correlation (see the
+  // `backups` table's updateId FK); operators don't read it during
+  // normal use, only when debugging a specific run via --json or
+  // `celilo backup list`. Surfacing it in human output added noise
+  // without paying for itself.
+  lines.push(`System update ${result.ok ? 'completed' : 'FAILED'}`);
   lines.push(`  audit verdict: ${result.audit.verdict}`);
   lines.push(
     `  self-update: ${result.selfUpdate.performed ? `${result.selfUpdate.from} → ${result.selfUpdate.to}` : `(${result.selfUpdate.reason})`}`,
@@ -500,6 +528,7 @@ export async function handleSystemUpdate(
     capabilityAbi: {
       modules: upgradableModules.map((m) => ({
         id: m.id,
+        state: m.state,
         manifest: m.manifestData as ModuleManifest,
       })),
     },
@@ -523,6 +552,7 @@ export async function handleSystemUpdate(
     moduleConfigs: {
       modules: upgradableModules.map((m) => ({
         id: m.id,
+        state: m.state,
         manifest: m.manifestData as ModuleManifest,
         configs: configsByModule.get(m.id) ?? {},
       })),
@@ -531,6 +561,7 @@ export async function handleSystemUpdate(
     backups: {
       modules: upgradableModules.map((m) => ({
         id: m.id,
+        state: m.state,
         manifest: m.manifestData as ModuleManifest,
         lastSuccessfulBackupAt: latestBackupByModule.get(m.id) ?? null,
       })),
@@ -587,14 +618,21 @@ export async function handleSystemUpdate(
   }
 
   // Decide whether the celilo-db snapshot is even needed for this run.
-  // If nothing's changing at the module level, there's nothing to roll
-  // back to — taking a snapshot would be pointless work, and on a fresh
-  // box (no storage configured yet) it would actively fail. Treat the
-  // "nothing-to-update" case as implicit --no-backup.
-  const hasModuleUpdates = [...snapshots.values()].some(
-    (s) => s.latestVersion && s.latestVersion !== s.installedVersion,
-  );
-  const effectiveNoBackup = noBackup || !hasModuleUpdates;
+  //
+  // The snapshot is a safety net — its purpose is to let `system update`
+  // roll back if a deploy / health step bricks a running module. So
+  // we ONLY need it when the run will actually upgrade-and-redeploy
+  // a currently-deployed module. Two cases that don't qualify:
+  //
+  //   1. Nothing has new code waiting at all (everyone's at latest).
+  //   2. The only modules with new code are IMPORTED-but-not-deployed.
+  //      Their upgrade is a pure source-files-on-disk refresh; no
+  //      live state to roll back, no risk to mitigate. Forcing a
+  //      backup here means an operator on a fresh celilo-mgmt with
+  //      nothing deployed yet has to configure backup storage just
+  //      to refresh the on-disk modules they imported — which is
+  //      exactly the friction that prompted this code path.
+  const effectiveNoBackup = noBackup || !shouldTakeBackup({ snapshots, wasDeployed });
 
   // Pre-flight the storage check so a missing/unusable default doesn't
   // reach the orchestrator's snapshot hook (where the throw would

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

@@ -24,7 +24,10 @@ function makeModule(
     hooks: opts.hasBackupHook ? { on_backup: { script: './backup.ts' } } : {},
     backup: opts.schedule ? { schedule: opts.schedule } : undefined,
   } as unknown as ModuleManifest;
-  return { id, manifest, lastSuccessfulBackupAt: opts.lastSuccessfulBackupAt };
+  // Default to INSTALLED — existing tests assert backup findings
+  // fire, which is the deployed-module behavior. Tests for the
+  // non-deployed-skip behavior override this explicitly.
+  return { id, state: 'INSTALLED', manifest, lastSuccessfulBackupAt: opts.lastSuccessfulBackupAt };
 }
 
 describe('auditBackups', () => {
@@ -230,4 +233,24 @@ describe('auditBackups', () => {
     expect(result).toHaveLength(2);
     expect(result.map((f) => f.subject).sort()).toEqual(['authentik', 'homebridge']);
   });
+
+  // The regression: backups were complaining about IMPORTED modules
+  // having no successful backup. There's no live state on an
+  // IMPORTED module, so there's nothing to back up — skip entirely
+  // rather than emit a misleading finding.
+  test('skips non-deployed (IMPORTED) modules entirely', async () => {
+    const importedModule = {
+      ...makeModule('authentik', {
+        hasBackupHook: true,
+        lastSuccessfulBackupAt: null,
+        schedule: 'daily',
+      }),
+      state: 'IMPORTED',
+    };
+    const result = await auditBackups({
+      modules: [importedModule],
+      now: () => NOW,
+    });
+    expect(result).toEqual([]);
+  });
 });

package/src/services/audit/backups.ts

@@ -20,6 +20,8 @@ import type { DriftFinding } from './types';
 
 export interface InstalledModuleBackupInfo {
   id: string;
+  /** Lifecycle state — non-deployed modules have nothing to back up. */
+  state: string;
   manifest: ModuleManifest;
   /** Most recent successful backup timestamp (ms since epoch), or null. */
   lastSuccessfulBackupAt: number | null;
@@ -87,12 +89,20 @@ function formatAge(ms: number): string {
   return `${mins}m`;
 }
 
+const DEPLOYED_STATES = new Set(['INSTALLED', 'VERIFIED']);
+
 export async function auditBackups(deps: BackupsAuditDeps): Promise<DriftFinding[]> {
   const now = (deps.now ?? Date.now)();
   const findings: DriftFinding[] = [];
 
   for (const m of deps.modules) {
     if (!moduleHasBackupHook(m.manifest)) continue;
+    // Non-deployed modules have no live state to back up. Surfacing
+    // a "no successful backup recorded" finding for an IMPORTED
+    // module is just noise — the operator hasn't deployed yet, so
+    // there's nothing to lose. Skip them entirely from the backup
+    // audit.
+    if (!DEPLOYED_STATES.has(m.state)) continue;
 
     if (m.lastSuccessfulBackupAt === null) {
       findings.push({

package/src/services/audit/capability-abi.test.ts

@@ -23,7 +23,10 @@ function makeModule(
     requires: opts.requires ? { capabilities: opts.requires } : { capabilities: [] },
     optional: opts.optional ? { capabilities: opts.optional } : undefined,
   } as unknown as ModuleManifest;
-  return { id, manifest };
+  // Default INSTALLED so existing assertions keep their `blocked`
+  // semantics. Tests that exercise the IMPORTED-demotion path
+  // override the state explicitly.
+  return { id, state: 'INSTALLED', manifest };
 }
 
 const RUNTIME = {

package/src/services/audit/capability-abi.ts

@@ -26,9 +26,19 @@ import type { DriftFinding } from './types';
 
 export interface InstalledCapabilityModule {
   id: string;
+  /**
+   * Lifecycle state — used to demote ABI mismatches on IMPORTED-but-
+   * not-deployed modules from `blocked` to `todo`. The mismatch is
+   * still real (the operator will hit it at deploy time) but it
+   * doesn't block other module work, and gating system update on it
+   * makes refreshing unrelated modules harder than it should be.
+   */
+  state: string;
   manifest: ModuleManifest;
 }
 
+const DEPLOYED_STATES = new Set(['INSTALLED', 'VERIFIED']);
+
 export interface CapabilityAbiAuditDeps {
   modules: InstalledCapabilityModule[];
   /** Override the framework registry — for tests. Defaults to `CAPABILITY_CONTRACT_VERSIONS`. */
@@ -120,7 +130,11 @@ export async function auditCapabilityAbi(deps: CapabilityAbiAuditDeps): Promise<
 
       findings.push({
         category: 'capability_abi',
-        severity: 'blocked',
+        // Demoted to `todo` for non-deployed modules: the mismatch
+        // matters at deploy time, but blocking system update on it
+        // means the operator can't refresh unrelated modules until
+        // they fix the ABI for a module they haven't deployed yet.
+        severity: DEPLOYED_STATES.has(m.state) ? 'blocked' : 'todo',
         code: 'capability_abi_provider_mismatch',
         message: `${m.id} provides ${p.name}@${p.version} but framework runtime expects ${runtimeVersion}`,
         details,
@@ -190,7 +204,9 @@ export async function auditCapabilityAbi(deps: CapabilityAbiAuditDeps): Promise<
 
       findings.push({
         category: 'capability_abi',
-        severity: 'blocked',
+        // Same demotion as the provider check — non-deployed
+        // modules surface as todos rather than blockers.
+        severity: DEPLOYED_STATES.has(m.state) ? 'blocked' : 'todo',
         code: 'capability_abi_consumer_mismatch',
         message: `${m.id} requires ${need.name}@${need.version} but ${provider.moduleId} provides ${provider.version}`,
         details,

package/src/services/audit/module-configs.test.ts

@@ -24,7 +24,10 @@ function makeModule(
     celilo_contract: '1.0',
     variables: { owns: variables, imports: [] },
   } as unknown as ModuleManifest;
-  return { id, manifest, configs };
+  // Default state INSTALLED so existing tests continue to expect
+  // `blocked` severity. Tests that exercise the IMPORTED path
+  // override the state explicitly.
+  return { id, state: 'INSTALLED', manifest, configs };
 }
 
 describe('auditModuleConfigs', () => {
@@ -128,4 +131,47 @@ describe('auditModuleConfigs', () => {
     expect(result).toHaveLength(1);
     expect(result[0].subject).toBe('caddy');
   });
+
+  // The exact regression the operator hit on celilo-mgmt: required
+  // configs missing on IMPORTED-but-not-yet-deployed modules used to
+  // surface as `blocked`, gating system update entirely. The deploy
+  // interview collects them at deploy time, so they're todos here.
+  test('IMPORTED module with missing required config → todo, not blocked', async () => {
+    const moduleWithImportedState = {
+      ...makeModule('namecheap', [makeVariable({ name: 'domains', required: true })], {}),
+      state: 'IMPORTED',
+    };
+    const result = await auditModuleConfigs({
+      modules: [moduleWithImportedState],
+    });
+    expect(result).toHaveLength(1);
+    expect(result[0].severity).toBe('todo');
+    // Remediation also flips: don't tell the operator to manually
+    // `module config set` (the deploy interview is the right path).
+    expect(result[0].remediation).toBe('celilo module deploy namecheap');
+  });
+
+  test('VALIDATED / GENERATING / other pre-deploy states also demote to todo', async () => {
+    const validated = {
+      ...makeModule('foo', [makeVariable({ name: 'x', required: true })], {}),
+      state: 'VALIDATED',
+    };
+    const generating = {
+      ...makeModule('bar', [makeVariable({ name: 'x', required: true })], {}),
+      state: 'GENERATING',
+    };
+    const result = await auditModuleConfigs({ modules: [validated, generating] });
+    expect(result).toHaveLength(2);
+    expect(result.every((f) => f.severity === 'todo')).toBe(true);
+  });
+
+  test('VERIFIED state stays blocked (deployed module)', async () => {
+    const verified = {
+      ...makeModule('caddy', [makeVariable({ name: 'acme_email', required: true })], {}),
+      state: 'VERIFIED',
+    };
+    const result = await auditModuleConfigs({ modules: [verified] });
+    expect(result).toHaveLength(1);
+    expect(result[0].severity).toBe('blocked');
+  });
 });

package/src/services/audit/module-configs.ts

@@ -8,20 +8,30 @@
  * the user.
  *
  * Severity rules:
- * - `required: true` AND no current value AND no `default:` → BLOCKED
- *   (the user has to set it; `system update` would otherwise fail at
- *   deploy time anyway).
- * - All other unset cases → no finding. An optional variable with a
- *   default is by definition fine when unset (the default applies);
- *   a required variable with a default is also fine (the default
- *   resolves the value). Either case as drift would just be noise.
+ * - DEPLOYED module (state=INSTALLED|VERIFIED) with `required: true`
+ *   AND no current value AND no `default:` → BLOCKED. The module is
+ *   live and a config gap is a real divergence: the next deploy
+ *   would fail, capability consumers may be reading the unset value,
+ *   etc.
+ * - NON-DEPLOYED module (IMPORTED, etc.) → TODO. The deploy
+ *   interview will collect required values when the operator
+ *   eventually runs `module deploy`. Telling them to manually
+ *   `module config set` is bad UX — the interview is the canonical
+ *   path. Surfacing as TODO keeps the visibility without escalating
+ *   the verdict.
+ * - Variable has a `default:` (any state) → no finding. The default
+ *   resolves the value.
  */
 
 import type { ModuleManifest, VariableDeclare } from '../../manifest/schema';
 import type { DriftFinding } from './types';
 
+const DEPLOYED_STATES = new Set(['INSTALLED', 'VERIFIED']);
+
 export interface InstalledModuleConfig {
   id: string;
+  /** Lifecycle state from the modules table — drives severity. */
+  state: string;
   manifest: ModuleManifest;
   /** Map of config key → current value (string for primitives, parsed object for complex). */
   configs: Record<string, unknown>;
@@ -37,6 +47,7 @@ function isUnset(value: unknown): boolean {
 
 function ownVariableFindings(
   moduleId: string,
+  state: string,
   variable: VariableDeclare,
   currentValue: unknown,
 ): DriftFinding[] {
@@ -47,23 +58,30 @@ function ownVariableFindings(
   if (!isUnset(currentValue)) return [];
 
   const hasDefault = variable.default !== undefined && variable.default !== null;
+  if (!variable.required || hasDefault) return [];
 
-  if (variable.required && !hasDefault) {
-    return [
-      {
-        category: 'module_configs',
-        severity: 'blocked',
-        code: 'module_config_required_unset',
-        message: `${moduleId}: required config "${variable.name}" is not set`,
-        details: variable.description,
-        remediation: `celilo module config set ${moduleId} ${variable.name} <value>`,
-        actionable: true,
-        subject: moduleId,
-      },
-    ];
-  }
+  // BLOCKED only when the module is currently deployed. For
+  // IMPORTED-and-similar pre-deploy states, the deploy interview
+  // collects this value automatically — surfacing as a blocker
+  // wrongly directs the operator at `module config set` (a manual
+  // workaround) when the right command is `module deploy <id>`.
+  const severity = DEPLOYED_STATES.has(state) ? 'blocked' : 'todo';
+  const remediation = DEPLOYED_STATES.has(state)
+    ? `celilo module config set ${moduleId} ${variable.name} <value>`
+    : `celilo module deploy ${moduleId}`;
 
-  return [];
+  return [
+    {
+      category: 'module_configs',
+      severity,
+      code: 'module_config_required_unset',
+      message: `${moduleId}: required config "${variable.name}" is not set`,
+      details: variable.description,
+      remediation,
+      actionable: true,
+      subject: moduleId,
+    },
+  ];
 }
 
 export async function auditModuleConfigs(deps: ModuleConfigsAuditDeps): Promise<DriftFinding[]> {
@@ -72,7 +90,7 @@ export async function auditModuleConfigs(deps: ModuleConfigsAuditDeps): Promise<
   for (const m of deps.modules) {
     const owned = m.manifest.variables?.owns ?? [];
     for (const variable of owned) {
-      findings.push(...ownVariableFindings(m.id, variable, m.configs[variable.name]));
+      findings.push(...ownVariableFindings(m.id, m.state, variable, m.configs[variable.name]));
     }
   }