@mycadet/assistant 0.12.7 → 0.12.9

package/src/config/bundled-skills/skill-management/SKILL.md

@@ -31,7 +31,7 @@ Do NOT use this skill when the user just wants to run an existing skill. That is
 - **Edit** an existing skill by scaffolding over it (replaces the body and needs `activation_hints` restated; every other frontmatter field you leave out keeps its current value, and an empty value clears one)
 - **Delete** an existing managed skill directory
 
-Skills created via `scaffold_managed_skill` become available for `skill_load` when a valid top-level `SKILL.md` is written under the skill directory.
+Skills created via `scaffold_cloud_skill` become available for `skill_load` when a valid top-level `SKILL.md` is written under the skill directory.
 
 ## Step 1 - Align with the user before building
 
@@ -100,7 +100,7 @@ Each completion criterion must bind a **tool call** to the **user-visible artifa
 ```markdown
 ## SKILL COMPLETE WHEN
 
-- [ ] `scaffold_managed_skill` wrote the SKILL.md and returned its path
+- [ ] `scaffold_cloud_skill` wrote the SKILL.md and returned its path
 - [ ] User confirmed the skill loads via `skill_load`
 - [ ] User saw the trigger phrases that will activate it
 ```
@@ -111,7 +111,7 @@ Each completion criterion must bind a **tool call** to the **user-visible artifa
 
 Past 500 lines the model loses things in the middle. Warnings get buried, branching loses visibility, and the file fights the task for the same context budget. If a skill is growing past 500 lines, split reference material into separate files the skill points to.
 
-Companion files ship through `scaffold_managed_skill`'s `files` input and live inside the skill folder:
+Companion files ship through `scaffold_cloud_skill`'s `files` input and live inside the skill folder:
 
 - **`references/*.md`** for failure modes, gotchas, and cached values the body should point to.
 - **`scripts/*`** for reusable code the procedure runs. Store the exact version that already ran successfully: pass `copy_from` with the tested file's absolute path instead of pasting its contents into `content`, so the bytes that shipped are the bytes that ran. Have the new skill's body invoke it through the baseDir placeholder (the word baseDir in curly braces), which resolves to that skill's folder when it loads. The terminal does not run from the skill folder, so a bare `scripts/...` path would fail. (The placeholder is spelled out here rather than written literally because this very body undergoes the same substitution.)

package/src/config/bundled-skills/skill-management/TOOLS.json

@@ -2,8 +2,8 @@
   "version": 1,
   "tools": [
     {
-      "name": "scaffold_managed_skill",
-      "description": "Create or update a managed skill in {workspaceDir}/skills. The skill becomes available for skill_load when a valid top-level SKILL.md is written under the skill directory. Never persist a skill without explicit user consent. Before persisting, test the snippet: write to a temp file with bash and run with `bun run /tmp/cadet-eval/snippet.ts`. Iterate up to 3 attempts, then ask the user. Clean up temp files after. Do not use file_write for temp files outside the working directory.",
+      "name": "scaffold_cloud_skill",
+      "description": "Create or update a cloud skill in {workspaceDir}/skills. The skill becomes available for skill_load when a valid top-level SKILL.md is written under the skill directory. Never persist a skill without explicit user consent. Before persisting, test the snippet: write to a temp file with bash and run with `bun run /tmp/cadet-eval/snippet.ts`. Iterate up to 3 attempts, then ask the user. Clean up temp files after. Do not use file_write for temp files outside the working directory.",
       "category": "skills",
       "risk": "high",
       "input_schema": {
@@ -91,12 +91,12 @@
           "activation_hints"
         ]
       },
-      "executor": "tools/scaffold-managed.ts",
+      "executor": "tools/scaffold-cloud.ts",
       "execution_target": "host"
     },
     {
-      "name": "delete_managed_skill",
-      "description": "Delete a managed skill directory from {workspaceDir}/skills. Never delete a skill without explicit user confirmation.",
+      "name": "delete_cloud_skill",
+      "description": "Delete a cloud skill directory from {workspaceDir}/skills. Never delete a skill without explicit user confirmation.",
       "category": "skills",
       "risk": "high",
       "input_schema": {
@@ -104,7 +104,7 @@
         "properties": {
           "skill_id": {
             "type": "string",
-            "description": "The ID of the managed skill to delete."
+            "description": "The ID of the cloud skill to delete."
           },
           "remove_from_index": {
             "type": "boolean",
@@ -113,12 +113,12 @@
         },
         "required": ["skill_id"]
       },
-      "executor": "tools/delete-managed.ts",
+      "executor": "tools/delete-cloud.ts",
       "execution_target": "host"
     },
     {
       "name": "find_similar_skills",
-      "description": "Find the existing skills most similar to a goal. Scores the goal against the skill catalog's capability pages and returns a ranked shortlist of nearest skills, each with its name, description, source (bundled, managed, plugin, workspace, or extra), and similarity score. For a managed hit it also returns `author`: \"assistant\" if the assistant authored that skill (so it may be overwritten) or \"user\" if a person did (off-limits); `author` is omitted for non-managed sources and for managed skills with no recorded author. For a background memory pass, a hit it may refine (managed, assistant-authored) also carries `current`: the skill as it is now, in scaffold_managed_skill's own argument names (name, description, emoji, category, includes, activation_hints, avoid_when, body_markdown), so a refinement can be written from the real skill and carry forward what it is not changing. Read-only: it never creates, edits, or deletes anything.",
+      "description": "Find the existing skills most similar to a goal. Scores the goal against the skill catalog's capability pages and returns a ranked shortlist of nearest skills, each with its name, description, source (bundled, managed, plugin, workspace, or extra), and similarity score. For a cloud hit it also returns `author`: \"assistant\" if the assistant authored that skill (so it may be overwritten) or \"user\" if a person did (off-limits); `author` is omitted for non-managed sources and for managed skills with no recorded author. For a background memory pass, a hit it may refine (managed, assistant-authored) also carries `current`: the skill as it is now, in scaffold_cloud_skill's own argument names (name, description, emoji, category, includes, activation_hints, avoid_when, body_markdown), so a refinement can be written from the real skill and carry forward what it is not changing. Read-only: it never creates, edits, or deletes anything.",
       "category": "skills",
       "risk": "low",
       "input_schema": {

package/src/config/bundled-skills/skill-management/tools/{delete-managed.ts → delete-cloud.ts}

@@ -1,4 +1,4 @@
-import { executeDeleteManagedSkill } from "../../../../tools/skills/delete-managed.js";
+import { executeDeleteCloudSkill } from "../../../../tools/skills/delete-cloud.js";
 import type {
   ToolContext,
   ToolExecutionResult,
@@ -8,5 +8,5 @@ export async function run(
   input: Record<string, unknown>,
   context: ToolContext,
 ): Promise<ToolExecutionResult> {
-  return executeDeleteManagedSkill(input, context);
+  return executeDeleteCloudSkill(input, context);
 }

package/src/config/bundled-skills/skill-management/tools/{scaffold-managed.ts → scaffold-cloud.ts}

@@ -1,4 +1,4 @@
-import { executeScaffoldManagedSkill } from "../../../../tools/skills/scaffold-managed.js";
+import { executeScaffoldCloudSkill } from "../../../../tools/skills/scaffold-cloud.js";
 import type {
   ToolContext,
   ToolExecutionResult,
@@ -8,5 +8,5 @@ export async function run(
   input: Record<string, unknown>,
   context: ToolContext,
 ): Promise<ToolExecutionResult> {
-  return executeScaffoldManagedSkill(input, context);
+  return executeScaffoldCloudSkill(input, context);
 }

package/src/config/call-site-defaults.ts

@@ -12,7 +12,7 @@ type CallSiteDefaultConfig = {
   /**
    * Direct model pin overriding the resolved profile's model. The resolver
    * treats a bare model pin as catalog-implied: it stamps the model's
-   * catalog provider and keeps the provider-agnostic Cadet managed
+   * catalog provider and keeps the provider-agnostic Cadet Cloud
    * connection (see `resolveOverrideOrDefault`). Reserve for call sites
    * whose latency envelope the profile's model cannot meet.
    */
@@ -89,8 +89,8 @@ export const CALL_SITE_DEFAULTS: Record<LLMCallSite, CallSiteDefaultConfig> = {
   // Vision captioning for the image-fallback plugin. No pinned profile — the
   // plugin resolves a vision-capable profile itself via `doesSupportVision` and
   // passes it as an `overrideProfile`, so the call-site default is a fallback
-  // that inherits the workspace default. Pinning a managed profile would break
-  // BYOK installs where managed profiles are uncredentialed.
+  // that inherits the workspace default. Pinning a cloud profile would break
+  // BYOK installs where cloud profiles are uncredentialed.
   vision: {},
 
   heartbeatAgent: {
@@ -153,14 +153,14 @@ export const CALL_SITE_DEFAULTS: Record<LLMCallSite, CallSiteDefaultConfig> = {
     effort: "low",
     thinking: { enabled: false },
   },
-  // The managed-only profile resolves through the Cadet connection. On a
+  // The cloud-only profile resolves through the Cadet connection. On a
   // BYOK install it has no body, so the judge is unavailable and the front
   // door's own decision stands.
   voiceEscalationJudge: {
     profile: JEV_MANAGED_PROFILE_KEY,
   },
   // Picks the default profile for a turn pinned to the Auto profile. On a
-  // BYOK install the managed-only Jev profile has no body, so the router is
+  // BYOK install the cloud-only Jev profile has no body, so the router is
   // unavailable and the turn runs on Balanced.
   autoProfileRouter: {
     profile: JEV_MANAGED_PROFILE_KEY,
@@ -205,7 +205,7 @@ export const CALL_SITE_DEFAULTS: Record<LLMCallSite, CallSiteDefaultConfig> = {
   },
   // Anonymous and schema leaves inherit the workspace default config (no pinned
   // profile) so they always resolve to a credentialed provider. Pinning a
-  // managed profile like `cost-optimized` breaks BYOK installs where the managed
+  // cloud profile like `cost-optimized` breaks BYOK installs where the cloud
   // profiles are uncredentialed. A per-leaf `profile` option or a `workflowLeaf`
   // call-site override still takes precedence for cost control.
   workflowLeaf: {

package/src/config/{platform-identity.ts → cloud-identity.ts}

@@ -4,7 +4,7 @@
  * Django maps an assistant API key to these ids via
  * `POST /v1/internal/assistants/validate/`. Process startup rehydrates from
  * that endpoint into the in-memory overrides. Resolution reads those
- * overrides (and `PLATFORM_ORGANIZATION_ID` / `PLATFORM_USER_ID` when set).
+ * overrides (and `CLOUD_ORGANIZATION_ID` / `CLOUD_USER_ID` when set).
  * When the in-memory assistant id is empty, the next resolve retries
  * validate (single-flight, with a cooldown after a failed attempt that holds
  * only while the same API key and base URL are stored).
@@ -14,24 +14,23 @@ import { credentialKey } from "../security/credential-key.js";
 import { getSecureKeyAsync } from "../security/secure-keys.js";
 import { getLogger } from "../util/logger.js";
 import {
-  getPlatformAssistantId,
-  getPlatformBaseUrl,
-  getPlatformOrganizationId,
-  getPlatformUserId,
-  setPlatformAssistantId,
-  setPlatformOrganizationId,
-  setPlatformUserId,
+  getCloudAssistantId,
+  getCloudBaseUrl,
+  getCloudOrganizationId,
+  getCloudUserId,
+  setCloudAssistantId,
+  setCloudOrganizationId,
+  setCloudUserId,
 } from "./env.js";
 
-const log = getLogger("platform-identity");
+const log = getLogger("cloud-identity");
 
-export const PLATFORM_IDENTITY_VALIDATE_PATH =
-  "/v1/internal/assistants/validate/";
+export const CLOUD_IDENTITY_VALIDATE_PATH = "/v1/internal/assistants/validate/";
 
 const VALIDATE_TIMEOUT_MS = 5_000;
 const ENSURE_COOLDOWN_MS = 10_000;
 
-export type PlatformIdentityIds = {
+export type CloudIdentityIds = {
   assistantId: string;
   organizationId: string;
   userId: string;
@@ -45,29 +44,29 @@ function asNonEmptyString(value: unknown): string | undefined {
   return trimmed.length > 0 ? trimmed : undefined;
 }
 
-export function applyPlatformIdentityIds(ids: PlatformIdentityIds): void {
+export function applyCloudIdentityIds(ids: CloudIdentityIds): void {
   if (ids.assistantId) {
-    setPlatformAssistantId(ids.assistantId);
+    setCloudAssistantId(ids.assistantId);
   }
   if (ids.organizationId) {
-    setPlatformOrganizationId(ids.organizationId);
+    setCloudOrganizationId(ids.organizationId);
   }
   if (ids.userId) {
-    setPlatformUserId(ids.userId);
+    setCloudUserId(ids.userId);
   }
 }
 
 /**
  * Trade an assistant API key for the bound platform ids.
  *
- * Returns null when the platform is unreachable, refuses the key, or the
+ * Returns null when the cloud is unreachable, refuses the key, or the
  * body has no ids. Never throws.
  */
-export async function fetchPlatformIdentityIds(
+export async function fetchCloudIdentityIds(
   baseUrl: string,
   apiKey: string,
-): Promise<PlatformIdentityIds | null> {
-  const url = `${baseUrl.replace(/\/+$/, "")}${PLATFORM_IDENTITY_VALIDATE_PATH}`;
+): Promise<CloudIdentityIds | null> {
+  const url = `${baseUrl.replace(/\/+$/, "")}${CLOUD_IDENTITY_VALIDATE_PATH}`;
   try {
     const res = await fetch(url, {
       method: "POST",
@@ -132,7 +131,7 @@ function identityFingerprint(apiKey: string, baseUrl: string): string {
   return `${baseUrl}\0${apiKey}`;
 }
 
-export function _resetPlatformIdentityEnsureForTests(): void {
+export function _resetCloudIdentityEnsureForTests(): void {
   ensureInFlight = null;
   failedFingerprint = null;
   nextEnsureAttemptAt = 0;
@@ -146,15 +145,15 @@ export function _resetPlatformIdentityEnsureForTests(): void {
  * attempt used and that attempt is still inside the cooldown window.
  * Concurrent callers share one in-flight request.
  */
-export async function ensurePlatformIdentityIds(): Promise<void> {
-  if (getPlatformAssistantId()?.trim()) {
+export async function ensureCloudIdentityIds(): Promise<void> {
+  if (getCloudAssistantId()?.trim()) {
     return;
   }
   if (!ensureInFlight) {
     ensureInFlight = (async () => {
       try {
         const apiKey = await readAssistantApiKey();
-        const baseUrl = (getPlatformBaseUrl() ?? "").replace(/\/+$/, "");
+        const baseUrl = (getCloudBaseUrl() ?? "").replace(/\/+$/, "");
         if (!apiKey || !baseUrl) {
           return;
         }
@@ -165,13 +164,13 @@ export async function ensurePlatformIdentityIds(): Promise<void> {
         ) {
           return;
         }
-        const ids = await fetchPlatformIdentityIds(baseUrl, apiKey);
+        const ids = await fetchCloudIdentityIds(baseUrl, apiKey);
         if (!ids) {
           failedFingerprint = fingerprint;
           nextEnsureAttemptAt = Date.now() + ENSURE_COOLDOWN_MS;
           return;
         }
-        applyPlatformIdentityIds(ids);
+        applyCloudIdentityIds(ids);
         failedFingerprint = null;
         nextEnsureAttemptAt = 0;
         log.info("Loaded platform identity from platform validate");
@@ -183,36 +182,34 @@ export async function ensurePlatformIdentityIds(): Promise<void> {
   await ensureInFlight;
 }
 
-export async function resolvePlatformAssistantId(): Promise<string> {
-  const existing = getPlatformAssistantId()?.trim() ?? "";
+export async function resolveCloudAssistantId(): Promise<string> {
+  const existing = getCloudAssistantId()?.trim() ?? "";
   if (existing) {
     return existing;
   }
-  await ensurePlatformIdentityIds();
-  return getPlatformAssistantId()?.trim() ?? "";
+  await ensureCloudIdentityIds();
+  return getCloudAssistantId()?.trim() ?? "";
 }
 
-export async function resolvePlatformAssistantIdOrNull(): Promise<
-  string | null
-> {
-  const id = await resolvePlatformAssistantId();
+export async function resolveCloudAssistantIdOrNull(): Promise<string | null> {
+  const id = await resolveCloudAssistantId();
   return id || null;
 }
 
-export async function resolvePlatformOrganizationId(): Promise<string> {
-  const existing = getPlatformOrganizationId()?.trim() ?? "";
+export async function resolveCloudOrganizationId(): Promise<string> {
+  const existing = getCloudOrganizationId()?.trim() ?? "";
   if (existing) {
     return existing;
   }
-  await ensurePlatformIdentityIds();
-  return getPlatformOrganizationId()?.trim() ?? "";
+  await ensureCloudIdentityIds();
+  return getCloudOrganizationId()?.trim() ?? "";
 }
 
-export async function resolvePlatformUserId(): Promise<string> {
-  const existing = getPlatformUserId()?.trim() ?? "";
+export async function resolveCloudUserId(): Promise<string> {
+  const existing = getCloudUserId()?.trim() ?? "";
   if (existing) {
     return existing;
   }
-  await ensurePlatformIdentityIds();
-  return getPlatformUserId()?.trim() ?? "";
+  await ensureCloudIdentityIds();
+  return getCloudUserId()?.trim() ?? "";
 }

package/src/config/cloud-rehydration.ts

@@ -0,0 +1,35 @@
+/**
+ * Rehydrate the in-memory cloud identity overrides at process startup.
+ *
+ * The identity overrides (the ID setters in `config/env.ts`) are normally
+ * only populated at runtime when the cloud pushes values. Any standalone
+ * process that talks to the cloud — the daemon and the schedule worker —
+ * must rehydrate them before its first request.
+ *
+ * Bound ids come from `POST /v1/internal/assistants/validate/`, which trades
+ * the assistant API key for assistant/organization/user ids. The base URL is
+ * intentionally NOT rehydrated here: it resolves from `config.cloud.baseUrl`
+ * or the `CADET_CLOUD_URL` environment variable, never the credential
+ * store.
+ *
+ * Safe to call more than once.
+ */
+import { getLogger } from "../util/logger.js";
+import { ensureCloudIdentityIds } from "./cloud-identity.js";
+
+const log = getLogger("cloud-rehydration");
+
+/**
+ * Rehydrate the cloud identity ids (assistant, organization, user). Safe to
+ * call more than once.
+ */
+export async function rehydrateCloudCredentials(): Promise<void> {
+  try {
+    await ensureCloudIdentityIds();
+  } catch (err) {
+    log.warn(
+      { error: err instanceof Error ? err.message : String(err) },
+      "Failed to rehydrate cloud identity ids (non-fatal)",
+    );
+  }
+}

package/src/config/{managed-speech-defaults.ts → cloud-speech-defaults.ts}

@@ -12,17 +12,17 @@
  * rule. It writes nothing, so runtime paths that only need to know which
  * provider they will actually use (live-voice readiness, transcription,
  * synthesis) resolve the same ids the scope-gated writer
- * {@link maybeDefaultSpeechToManaged} would persist, without needing write
+ * {@link maybeDefaultSpeechToCloud} would persist, without needing write
  * access to the config.
  */
 
 import { ttsSecretResolves } from "../calls/telephony-tts-capability.js";
-import { managedSpeechAvailable } from "../platform/managed-speech.js";
+import { managedSpeechAvailable } from "../cloud/cloud-speech.js";
 import { fluxModelForLanguage } from "../providers/speech-to-text/deepgram-flux-frames.js";
 import {
   baseModelFamilyFor,
   getProviderEntry,
-  isManagedSttProvider,
+  isCloudSttProvider,
   resolveSttCatalogKey,
   sttConfigForCatalogKey,
   supportsProviderTurnDetection,
@@ -48,7 +48,7 @@ import {
 } from "./loader.js";
 import type { AssistantConfig } from "./types.js";
 
-const log = getLogger("managed-speech-defaults");
+const log = getLogger("cloud-speech-defaults");
 
 /** Whether the configured BYOK STT provider has a usable credential. */
 async function sttByokCredentialResolves(provider: string): Promise<boolean> {
@@ -76,10 +76,10 @@ async function ttsByokCredentialsResolve(provider: string): Promise<boolean> {
 }
 
 /**
- * The managed provider that stands in for a configured one whose credential
+ * The cloud provider that stands in for a configured one whose credential
  * did not resolve.
  *
- * Managed speech has two STT entries, and they are not interchangeable:
+ * Cloud speech has two STT entries, and they are not interchangeable:
  * `cadet-flux` decides end-of-turn itself, `cadet` leaves it to the
  * session's silence timer. Substituting the plain one for a provider that had
  * provider turn detection would keep live voice working while quietly
@@ -100,7 +100,7 @@ function managedStandInFor(
   return fluxModelForLanguage(language) === null ? "cadet" : "cadet-flux";
 }
 
-/** The speech providers a runtime path uses, after managed-speech defaulting. */
+/** The speech providers a runtime path uses, after cloud-speech defaulting. */
 export interface EffectiveSpeechProviders {
   stt: SttProviderId;
   tts: TtsProviderId;
@@ -110,7 +110,7 @@ export interface EffectiveSpeechProviders {
  * Resolve the speech providers the runtime actually uses.
  *
  * A configured service whose BYOK credential does not resolve is reported as
- * its managed stand-in while managed speech is available (see
+ * its managed stand-in while cloud speech is available (see
  * {@link managedStandInFor}, which preserves provider turn detection); every
  * other service keeps its configured provider. Read-only: callers that hold no `settings.write`
  * scope (the live-voice WebSocket transport) resolve the same verdict the
@@ -134,7 +134,7 @@ export async function resolveEffectiveSpeechProviders(
   }
 
   const stt =
-    !isManagedSttProvider(configuredStt) &&
+    !isCloudSttProvider(configuredStt) &&
     !(await sttByokCredentialResolves(configuredStt))
       ? managedStandInFor(configuredStt, services.stt.language)
       : configuredStt;
@@ -233,9 +233,9 @@ function globalSttUpdates(selection: {
  *
  * Safe to call repeatedly (idempotent) and safe to fire-and-forget. Callers
  * must hold the `settings.write` scope — this is the only path that writes
- * `services.stt/tts.provider` on behalf of managed-speech defaulting.
+ * `services.stt/tts.provider` on behalf of cloud-speech defaulting.
  */
-export async function maybeDefaultSpeechToManaged(): Promise<void> {
+export async function maybeDefaultSpeechToCloud(): Promise<void> {
   try {
     const services = getConfig().services;
     const effective = await resolveEffectiveSpeechProviders();
@@ -266,7 +266,7 @@ export async function maybeDefaultSpeechToManaged(): Promise<void> {
           });
         }
         // The other consumers still need a provider that answers, and the
-        // narrow stand-in's own base is the managed row that serves them.
+        // narrow stand-in's own base is the cloud row that serves them.
         if (servesEveryRole(standIn.provider as SttProviderId)) {
           updates.push(...globalSttUpdates({ provider: standIn.provider }));
         }
@@ -294,7 +294,7 @@ export async function maybeDefaultSpeechToManaged(): Promise<void> {
     // Convenience defaulting must never break credential storage.
     log.warn(
       { error: err instanceof Error ? err.message : String(err) },
-      "Managed speech defaulting failed (non-fatal)",
+      "Cloud speech defaulting failed (non-fatal)",
     );
   }
 }