switchroom 0.15.45 → 0.16.5

package/dist/cli/drive-write-pretool.mjs

@@ -4114,7 +4114,8 @@ var init_protocol = __esm(() => {
     op: exports_external.literal("probe-quota"),
     id: exports_external.string().min(1),
     accounts: exports_external.array(exports_external.string().min(1)).min(1).max(32),
-    timeoutMs: exports_external.number().int().positive().max(60000).optional()
+    timeoutMs: exports_external.number().int().positive().max(60000).optional(),
+    forceLive: exports_external.boolean().optional()
   });
   ClaimNotificationRequestSchema = exports_external.object({
     v: exports_external.literal(PROTOCOL_VERSION),
@@ -4165,7 +4166,8 @@ var init_protocol = __esm(() => {
     fallback_order: exports_external.array(exports_external.string()),
     accounts: exports_external.array(AccountStateSchema),
     agents: exports_external.array(AgentStateSchema),
-    consumers: exports_external.array(ConsumerStateSchema)
+    consumers: exports_external.array(ConsumerStateSchema),
+    active_overage_serving: exports_external.boolean().optional()
   });
   SetActiveDataSchema = exports_external.object({
     active: exports_external.string(),
@@ -4344,13 +4346,14 @@ class AuthBrokerClient {
     });
     return data;
   }
-  async probeQuota(accounts, timeoutMs) {
+  async probeQuota(accounts, timeoutMs, forceLive) {
     const data = await this.send({
       v: PROTOCOL_VERSION,
       id: randomUUID(),
       op: "probe-quota",
       accounts: [...accounts],
-      ...timeoutMs !== undefined ? { timeoutMs } : {}
+      ...timeoutMs !== undefined ? { timeoutMs } : {},
+      ...forceLive ? { forceLive: true } : {}
     });
     const parsed = data;
     for (const entry of parsed.results) {

package/dist/cli/notion-write-pretool.mjs

@@ -11898,7 +11898,7 @@ var TelegramChannelSchema = exports_external.object({
   format: exports_external.enum(["html", "markdownv2", "text"]).optional().describe("Default reply format passed to the plugin"),
   rate_limit_ms: exports_external.number().optional().describe("Minimum delay between outgoing messages in ms"),
   stream_mode: exports_external.enum(["pty", "checklist"]).optional().describe("How live progress is streamed to Telegram during a turn. " + "'pty' (default) surfaces text snapshots of Claude Code's TUI \u2014 " + "compatible but can flicker as Ink re-renders. 'checklist' drives " + "a structured progress card from session-tail events \u2014 stable " + "order, per-tool status emojis, fires only on semantic transitions."),
-  stream_throttle_ms: exports_external.number().int().nonnegative().optional().describe("Throttle window in ms between successive stream edits (or " + "sendMessageDraft tics) during a turn. Lower = more responsive " + "stream, higher = fewer API calls. Floored at 250 by draft-stream " + "itself. Default 300 for draft transport (DMs) and 1000 for " + "message transport (groups/forums). Override per-agent if a " + "particular agent needs snappier or quieter streaming."),
+  stream_throttle_ms: exports_external.number().int().nonnegative().optional().describe("Throttle window in ms between successive in-place stream edits " + "during a turn. Lower = more responsive stream, higher = fewer API " + "calls. Floored at 250 by draft-stream itself. Default 400 ms for DMs " + "and 1000 ms for groups/forums (respects Telegram's ~1 edit/sec/message " + "practical ceiling). Override per-agent if a particular agent needs " + "snappier or quieter streaming."),
   clear_status_on_completion: exports_external.boolean().optional().describe("When true, the live activity/status feed (the in-place 'what it's " + "doing' message \u2014 Reading X, Searching the web for Y, \u2026) is DELETED " + "when the turn's final answer lands, so only the reply remains. " + "Default false: the status message is left in the chat as a record " + "(its last step marked done) \u2014 no post-then-delete. Per-agent " + "override; cascades defaults \u2192 profile \u2192 agent (per-key)."),
   hotReloadStable: exports_external.boolean().optional().describe("If true, the stable workspace prefix (AGENTS.md, SOUL.md, USER.md, " + "IDENTITY.md, TOOLS.md) is re-injected on every turn via " + "the UserPromptSubmit hook instead of baked into --append-system-prompt " + "at session start. Lets workspace edits propagate without a restart. " + "Costs ~5-10% per-turn latency/spend since the stable prefix is no " + "longer prompt-cached."),
   inject_on_change: exports_external.boolean().optional().describe("Context-efficiency gate for per-turn hook injection (default true). " + "When true (the default), the turn-pacing directive and dynamic " + "workspace content are only re-emitted when their content changes or " + "the session_id changes \u2014 suppressing redundant injection that " + "otherwise triples compaction frequency. Set to false to revert to " + "the legacy always-emit behaviour (every turn injects the full " + "content regardless of whether it changed)."),
@@ -11986,6 +11986,14 @@ var GoogleWorkspaceConfigSchema = exports_external.object({
   approvers: exports_external.array(ApproverIdSchema).min(1).describe("Array of numeric Telegram user IDs authorized to approve drive onboarding. " + "At least one must be specified."),
   tier: GoogleWorkspaceTierSchema.optional().describe("RFC G Phase 1: which upstream MCP tier to expose. " + "core (default) = ~16 tools (Drive+Docs+Sheets+Calendar). " + "extended = ~40 tools (+Slides, Forms, Tasks, Chat). " + "complete = ~60+ tools (+Gmail; not recommended yet \u2014 see RFC G \u00a75).")
 }).optional();
+var LiteLLMConfigSchema = exports_external.object({
+  enabled: exports_external.boolean().optional().describe("Opt-in toggle. When true, `switchroom apply` provisions a per-agent " + "LiteLLM virtual key and injects routing env into the container. " + "Default OFF."),
+  base_url: exports_external.string().optional().describe("LiteLLM proxy base URL the agent's claude CLI routes through, e.g. " + "'http://127.0.0.1:4010'. Agents use network_mode:host, so loopback " + "reaches a host-bound proxy. Exported as ANTHROPIC_BASE_URL."),
+  admin_key: exports_external.string().optional().describe("LiteLLM master/admin key used at apply time to provision the team + " + "virtual key. Supports a vault reference (e.g. " + "'vault:litellm/master-key') \u2014 resolution happens at apply time via " + "the vault-broker. Never injected into the agent container."),
+  team: exports_external.string().optional().describe("LiteLLM team alias the per-agent key is created under. Defaults to " + "'switchroom' (applied in code, not as a schema default)."),
+  small_fast_model: exports_external.string().optional().describe("Model id exported as ANTHROPIC_SMALL_FAST_MODEL for the claude CLI's " + "background/fast lane, e.g. 'claude-haiku-4-5-20251001'."),
+  tags: exports_external.record(exports_external.string(), exports_external.string()).optional().describe("Extra key/value metadata tags attached to the provisioned LiteLLM " + "virtual key. Merged per-key across cascade layers (agent wins).")
+}).optional().describe("LiteLLM routing config \u2014 opt-in per-agent virtual-key auto-provisioning " + "+ routing env. Default OFF. See LiteLLMConfigSchema doc for the full flow.");
 var MicrosoftWorkspaceConfigSchema = exports_external.object({
   microsoft_client_id: exports_external.string().min(1).optional().describe("Microsoft OAuth application (client) ID from Entra portal " + "(literal string or vault reference e.g. " + "'vault:microsoft-oauth-client-id'). OPTIONAL \u2014 omit it to use " + "switchroom's shipped default Microsoft app (zero-config). " + "Set it only to bring your own Entra app (BYO)."),
   microsoft_client_secret: exports_external.string().min(1).optional().describe("Microsoft OAuth client secret. Optional \u2014 public-client apps " + "(Mobile + Desktop platform with 'Allow public client flows' " + "enabled) work without a secret; confidential clients pass " + "one. Either literal or vault reference e.g. " + "'vault:microsoft-oauth-client-secret'."),
@@ -12082,6 +12090,7 @@ var profileFields = {
   mcp_servers: exports_external.record(exports_external.string(), exports_external.unknown()).optional(),
   hooks: AgentHooksSchema,
   env: exports_external.record(exports_external.string(), exports_external.string()).optional(),
+  litellm: LiteLLMConfigSchema,
   system_prompt_append: exports_external.string().optional(),
   skills: exports_external.array(exports_external.string()).optional(),
   bundled_skills: exports_external.record(exports_external.string(), exports_external.boolean()).optional().describe("Opt-out map for switchroom's bundled-default skills " + "(e.g. skill-creator, mcp-builder, webapp-testing, pdf, docx, " + "xlsx, pptx, switchroom-cli, switchroom-status, switchroom-health). " + "Set a key to `false` to suppress that default for this agent. " + "Cascades from defaults.bundled_skills."),
@@ -12154,6 +12163,7 @@ var AgentSchema = exports_external.object({
   mcp_servers: exports_external.record(exports_external.string(), exports_external.unknown()).optional().describe("Additional MCP server configurations"),
   hooks: AgentHooksSchema.describe("Claude Code lifecycle hooks (SessionStart, UserPromptSubmit, Stop, etc). " + "Written to settings.json.hooks in Claude Code's native shape."),
   env: exports_external.record(exports_external.string(), exports_external.string()).optional().describe("Environment variables exported in start.sh before claude runs"),
+  litellm: LiteLLMConfigSchema.describe("Per-agent LiteLLM routing override. Presence with `enabled: true` opts " + "this agent IN to per-agent virtual-key auto-provisioning + routing env " + "(falls back to the top-level `litellm:` block for base_url/admin_key/" + "team/small_fast_model). Deep-merges one level over defaults/profile; " + "`tags` merge per-key, agent wins. Default OFF."),
   system_prompt_append: exports_external.string().optional().describe("Text passed via claude's --append-system-prompt flag. " + "Appended to the default or CLAUDE.md-derived system prompt."),
   skills: exports_external.array(exports_external.string()).optional().describe("Names of skills from switchroom.skills_dir to symlink into this " + "agent's skills/ directory. Unioned with defaults.skills."),
   bundled_skills: exports_external.record(exports_external.string(), exports_external.boolean()).optional().describe("Per-agent override of switchroom's bundled-default skills " + "(skill-creator, mcp-builder, webapp-testing, pdf, docx, xlsx, " + "pptx, switchroom-cli/status/health). Set a key to `false` to " + "opt out for this agent. Per-agent value wins over defaults.bundled_skills."),
@@ -12279,7 +12289,7 @@ var WebServiceConfigSchema = exports_external.object({
 });
 var HostdConfigSchema = exports_external.object({
   config_edit_enabled: exports_external.boolean().default(false).describe("Opt-in toggle for the `config_propose_edit` hostd verb (RFC " + "admin-agent-config-edit \u00a73). Default false \u2014 the verb returns " + "`E_CONFIG_EDIT_DISABLED` until the operator explicitly flips " + "this to true. When true, admin agents can propose unified-diff " + "patches against " + "`/state/config/switchroom.yaml`, gated by an operator approval " + "card in the primary chat. Same trust posture as `update_apply` " + "and `agent_restart`: the human-in-the-loop tap is the security " + "boundary, not the agent's judgement."),
-  config_edit_rate_per_hour: exports_external.number().int().min(1).max(20).default(3).describe("Per-requesting-agent rate cap for `config_propose_edit` cards " + "(RFC admin-agent-config-edit \u00a75). Default 3 cards/hour; min 1, " + "max 20. Configurable now, but the rate limiter is not yet enforced " + "(no `E_RATE_LIMITED` is currently raised); the field is reserved so " + "operators can pin the cap ahead of the limiter going live.")
+  config_edit_rate_per_hour: exports_external.number().int().min(1).max(20).default(3).describe("Per-requesting-agent rate cap for `config_propose_edit` cards " + "(RFC admin-agent-config-edit \u00a75). Default 3 cards/hour; min 1, " + "max 20. ENFORCED server-side: a caller exceeding this in a sliding " + "1-hour window is rejected with `E_RATE_LIMITED` (carrying a " + "`retry_after` fix) instead of posting another operator approval " + "card \u2014 so a looping agent is throttled rather than spamming the chat.")
 });
 var CronEgressSchema = exports_external.object({
   allowed_hosts: exports_external.array(exports_external.string().min(1)).default([]).describe("Hosts a poll may reach (exact, https-only). loopback/private/IP-literal are always rejected."),
@@ -12312,11 +12322,14 @@ var SwitchroomConfigSchema = exports_external.object({
         message: "Consumer name must be a path-safe slug (letters, digits, underscore, hyphen)"
       }).describe("Socket-path identity; binds at /run/switchroom/auth-broker/<name>/sock"),
       account: exports_external.string().min(1).describe("Pinned account label for this consumer. `get-credentials` returns " + "this account's credentials; `mark-exhausted` from this consumer " + "only affects this account."),
-      uid: exports_external.number().int().nonnegative().optional().describe("Optional UID to chown the consumer socket to (defaults to 0 = root, " + "suitable for sibling containers running as root).")
-    })).optional().describe("Non-agent peers that hold a broker socket (RFC H \u00a74.8). Each gets " + "its own `/run/switchroom/auth-broker/<name>/sock` chowned to its UID. " + "Consumers cannot be admins; a consumer name that collides with an " + "agent (whether that agent has `admin: true` or not) is a config " + "error caught at schema validation.")
+      uid: exports_external.number().int().nonnegative().optional().describe("Optional UID to chown the consumer socket to (defaults to 0 = root, " + "suitable for sibling containers running as root)."),
+      mirror_dir: exports_external.string().optional().describe("Optional host-side directory path. When set, the broker actively " + "writes the consumer's effective-account `.credentials.json` mirror " + "here \u2014 in addition to serving creds on demand via `get-credentials`. " + "Use this to eliminate the pull-latency gap: without a mirror the " + "consumer only gets failover creds at its next scheduled re-fetch " + "(up to 30 min). With a mirror the broker pushes failover creds " + "immediately when it detects exhaustion (consumer-quota-sensor tick, " + "or a mark-exhausted RPC on the pinned account). The directory must " + "be accessible to the broker container (bind-mounted from the host) " + "and to the consumer container; the broker writes " + "`<mirror_dir>/.credentials.json` atomically. Chown is attempted to " + "`uid` (default 0) \u2014 swallowed when CAP_CHOWN is absent.")
+    })).optional().describe("Non-agent peers that hold a broker socket (RFC H \u00a74.8). Each gets " + "its own `/run/switchroom/auth-broker/<name>/sock` chowned to its UID. " + "Consumers cannot be admins; a consumer name that collides with an " + "agent (whether that agent has `admin: true` or not) is a config " + "error caught at schema validation."),
+    allow_overage_accounts: exports_external.array(exports_external.string().min(1)).optional().describe("Opt-in list of account labels (bare strings matching `auth.active` / " + "`auth.fallback_order` entries) that may be served PAST the weekly " + "utilization wall when Anthropic overage billing is available for the " + "account (`overageStatus === 'allowed'`). Overage is REAL MONEY \u2014 " + "default is empty (no account gets this). An account in this list is " + "only kept eligible when its fresh quota snapshot reports " + "`overageStatus: 'allowed'` AND `overageDisabledReason` is NOT " + "'out_of_credits' (i.e. the overage credit has not been exhausted). " + "As soon as `overageDisabledReason` becomes 'out_of_credits', the " + "account is blocked immediately regardless of this flag. Overage lifts " + "ONLY the utilization wall \u2014 it cannot lift an active exhaustion mark " + "written by a real 429 (`mark-exhausted`).")
   }).optional().describe("Switchroom-auth-broker configuration (RFC H). Fleet-wide active account, " + "fallback order, admin-agent ACL, and ephemeral-consumer surface. " + "Required from the v0.8+ schema onwards; pre-v0.8 fleets are migrated " + "in-place by `switchroom apply` (see src/auth/migrate-schema.ts)."),
   drive: GoogleWorkspaceConfigSchema.describe("RFC D legacy key \u2014 use `google_workspace:` instead. Optional Google " + "Workspace onboarding configuration. When set, supplies Google OAuth " + "client credentials, the approver allowlist for `switchroom drive " + "connect`, and the optional tier knob. Env vars " + "(SWITCHROOM_GOOGLE_CLIENT_ID, SWITCHROOM_GOOGLE_CLIENT_SECRET, " + "SWITCHROOM_APPROVER_USER_ID) take precedence over this block when " + "set, preserving back-compat with the env-only flow shipped in #766."),
   google_workspace: GoogleWorkspaceConfigSchema.describe("RFC G canonical key. Top-level Google Workspace configuration \u2014 " + "OAuth client credentials, approver allowlist, and tier knob (`core` " + "| `extended` | `complete`, default `core`). Mutually exclusive with " + "`drive:` at the top level (loader fails fast if both are set)."),
+  litellm: LiteLLMConfigSchema.describe("Top-level LiteLLM routing infra \u2014 global base_url, admin_key (the " + "LiteLLM master key, supports a `vault:` ref), team alias, and " + "small_fast_model shared by every agent that opts in. Set `enabled: " + "true` here to default the whole fleet on (each agent can still set " + "`litellm.enabled: false` to opt out). Default OFF."),
   microsoft_workspace: MicrosoftWorkspaceConfigSchema.describe("RFC #1873 (Microsoft 365 integration). Top-level Microsoft Workspace " + "configuration \u2014 OAuth client credentials (Entra app), authority " + "endpoint (defaults to /common for personal MSA + work), and the " + "org_mode opt-in for Teams/SharePoint surfaces. Block is optional; " + "when omitted the broker does not register the Microsoft provider."),
   notion_workspace: NotionWorkspaceConfigSchema.describe("RFC reference/rfcs/notion-integration.md. Top-level Notion integration " + "config \u2014 vault key for the integration token, friendly-name \u2192 " + "database UUID map, optional MCP-package version pin, and optional " + "global rate-limit override (default 3 rps, Notion's documented " + "public-API limit). Block is optional; when omitted no agent gets a " + "Notion MCP entry regardless of per-agent config."),
   quota: QuotaConfigSchema.optional().describe("Optional weekly/monthly USD spend budgets rendered in the session " + "greeting. Usage is read from ccusage at runtime; no network calls."),
@@ -12738,6 +12751,24 @@ function mergeAgentConfig(defaultsIn, agentIn) {
       ...merged.env ?? {}
     };
   }
+  if (defaults.litellm || merged.litellm) {
+    const base = defaults.litellm ?? {};
+    const override = merged.litellm ?? {};
+    const combined = { ...base };
+    for (const [k, v] of Object.entries(override)) {
+      if (v === undefined)
+        continue;
+      if (k === "tags" && base.tags && typeof v === "object" && v !== null && !Array.isArray(v)) {
+        combined.tags = {
+          ...base.tags,
+          ...v
+        };
+      } else {
+        combined[k] = v;
+      }
+    }
+    merged.litellm = combined;
+  }
   if (defaults.subagents || merged.subagents) {
     const dSub = defaults.subagents ?? {};
     const mSub = merged.subagents ?? {};