@autohq/cli 0.1.564 → 0.1.566

package/dist/index.js

@@ -23003,8 +23003,239 @@ var init_runtime_restart_diagnostics = __esm({
   }
 });
 
+// ../../packages/schemas/src/session-attribution-epochs.ts
+var ATTRIBUTION_GRADES, AttributionGradeSchema, AttributionRequesterSnapshotSchema, SessionAttributionEpochCreateSchema, SessionAttributionEpochRecordSchema;
+var init_session_attribution_epochs = __esm({
+  "../../packages/schemas/src/session-attribution-epochs.ts"() {
+    "use strict";
+    init_zod();
+    init_ids();
+    init_model_credential_funding();
+    init_requester();
+    ATTRIBUTION_GRADES = ["derived", "asserted"];
+    AttributionGradeSchema = external_exports.enum(ATTRIBUTION_GRADES);
+    AttributionRequesterSnapshotSchema = external_exports.object({
+      origin: RequesterRootOriginSchema,
+      userId: UserIdSchema.nullable()
+    }).strict();
+    SessionAttributionEpochCreateSchema = external_exports.object({
+      organizationId: OrganizationIdSchema,
+      projectId: ProjectIdSchema.nullable(),
+      sessionId: SessionIdSchema,
+      sourceCommandId: SessionCommandIdSchema,
+      requesterKey: external_exports.string().trim().min(1),
+      requesterSnapshot: AttributionRequesterSnapshotSchema.nullable(),
+      attributionGrade: AttributionGradeSchema,
+      assertedByCommandId: SessionCommandIdSchema.nullable(),
+      assertedBySessionId: SessionIdSchema.nullable(),
+      assertedByAgentResourceId: external_exports.string().trim().min(1).nullable(),
+      fundingSource: ModelCredentialFundingSourceSchema.nullable(),
+      providerGrantId: ProviderGrantIdSchema.nullable(),
+      userModelSubscriptionId: UserSubscriptionCredentialIdSchema.nullable()
+    }).strict();
+    SessionAttributionEpochRecordSchema = SessionAttributionEpochCreateSchema.extend({
+      id: SessionAttributionEpochIdSchema,
+      createdAt: external_exports.string().datetime()
+    }).strict();
+  }
+});
+
+// ../../packages/schemas/src/usage.ts
+var UsageHarnessSchema, NonNegativeTokenCountSchema, NonNegativeUsdSchema2, UsageEventSchema, UsageEventRecordSchema, UsageTotalsSchema, SessionUsageModelSummarySchema, UsageFundingSummarySchema, SessionModelFundingStateSchema, SessionUsageCurrentAttributionSchema, SessionUsageRequesterSummarySchema, SessionUsageResponseSchema, PROJECT_USAGE_RANGE_KEYS, ProjectUsageRangeKeySchema, UtcDateSchema, ProjectUsageTotalsSchema, ProjectUsageAgentSummarySchema, ProjectUsageModelSummarySchema, ProjectUsageSubscriptionSummarySchema, ProjectUsageSubscriptionSliceSchema, ProjectUsageRequesterAttributionSchema, ProjectUsageRequesterSummarySchema, ProjectUsageDayAgentSliceSchema, ProjectUsageDayModelSliceSchema, ProjectUsageDayPointSchema, ProjectUsageResponseSchema;
+var init_usage = __esm({
+  "../../packages/schemas/src/usage.ts"() {
+    "use strict";
+    init_zod();
+    init_agents();
+    init_ids();
+    init_model_credential_funding();
+    init_session_attribution_epochs();
+    init_model_credential_funding();
+    UsageHarnessSchema = external_exports.enum(AGENT_HARNESSES);
+    NonNegativeTokenCountSchema = external_exports.number().int().nonnegative();
+    NonNegativeUsdSchema2 = external_exports.number().nonnegative();
+    UsageEventSchema = external_exports.object({
+      // Tenant + lineage attribution. projectId is nullable because a resource can
+      // belong directly to an organization with no project.
+      organizationId: OrganizationIdSchema,
+      projectId: ProjectIdSchema.nullable(),
+      sessionId: SessionIdSchema,
+      // Authoritative platform turn association. Null/absent is an explicit
+      // compatibility state for legacy writers and usage outside a model turn;
+      // readers must never infer it from timestamps or ledger order.
+      turnId: SessionCommandIdSchema.nullable().optional(),
+      agentResourceId: external_exports.string().trim().min(1),
+      harness: UsageHarnessSchema,
+      // The concrete model id the call billed against (e.g. "claude-opus-4-8"),
+      // taken from the response rather than any selection layer — a single run can
+      // touch multiple models, so this is per-call, never per-run.
+      model: external_exports.string().trim().min(1),
+      inputTokens: NonNegativeTokenCountSchema,
+      outputTokens: NonNegativeTokenCountSchema,
+      cacheCreationTokens: NonNegativeTokenCountSchema,
+      cacheReadTokens: NonNegativeTokenCountSchema,
+      // The provider's own cost when it reports one, kept only for reconciliation.
+      // Nullable/absent because it can be 0 or missing — never the billing source.
+      providerCostUsd: NonNegativeUsdSchema2.nullable(),
+      // Platform-derived cost and the rate-card version it was derived under, so a
+      // later rate-card change never silently rewrites historical cost.
+      derivedCostUsd: NonNegativeUsdSchema2,
+      pricingVersion: external_exports.string().trim().min(1),
+      // What this call debits from the organization's credit balance, and the fee
+      // card it was derived under (see `./billing`). Required — a writer that
+      // forgets it would silently record unbilled usage, so there is no default.
+      billedCostUsd: NonNegativeUsdSchema2,
+      billingFeeVersion: external_exports.string().trim().min(1),
+      // Historical funding attribution for the model call. The provider grant id is
+      // denormalized and nullable for platform-funded usage, matching the ledger's
+      // no-FK audit-trail contract.
+      fundingSource: ModelCredentialFundingSourceSchema.default("platform"),
+      modelCredentialProviderGrantId: ProviderGrantIdSchema.nullable().default(null),
+      // Optional during the cross-chunk rollout: B1 defines the attribution field
+      // before B3 deploys its persistence column and D3 begins populating it.
+      modelCredentialUserSubscriptionId: UserSubscriptionCredentialIdSchema.nullable().optional(),
+      attributionEpochId: SessionAttributionEpochIdSchema.nullable().optional(),
+      requesterKey: external_exports.string().trim().min(1).nullable().optional(),
+      attributionGrade: AttributionGradeSchema.nullable().optional(),
+      occurredAt: external_exports.string().datetime()
+    });
+    UsageEventRecordSchema = UsageEventSchema.extend({
+      id: external_exports.string().trim().min(1),
+      turnId: SessionCommandIdSchema.nullable().default(null),
+      attributionEpochId: SessionAttributionEpochIdSchema.nullable().default(null),
+      requesterKey: external_exports.string().trim().min(1).nullable().default(null),
+      attributionGrade: AttributionGradeSchema.nullable().default(null),
+      // Set on append-only correction rows: this row replaces the pointed-at row's
+      // amounts, and ledger readers exclude any row another row supersedes.
+      supersedesUsageEventId: external_exports.string().trim().min(1).nullable().default(null),
+      createdAt: external_exports.string().datetime()
+    });
+    UsageTotalsSchema = external_exports.object({
+      eventCount: NonNegativeTokenCountSchema,
+      inputTokens: NonNegativeTokenCountSchema,
+      outputTokens: NonNegativeTokenCountSchema,
+      cacheCreationTokens: NonNegativeTokenCountSchema,
+      cacheReadTokens: NonNegativeTokenCountSchema,
+      totalTokens: NonNegativeTokenCountSchema,
+      billedCostUsd: NonNegativeUsdSchema2
+    });
+    SessionUsageModelSummarySchema = UsageTotalsSchema.extend({
+      model: external_exports.string().trim().min(1)
+    });
+    UsageFundingSummarySchema = UsageTotalsSchema.extend({
+      fundingSource: ModelCredentialFundingSourceSchema,
+      coveredInferenceCostUsd: NonNegativeUsdSchema2
+    });
+    SessionModelFundingStateSchema = external_exports.object({
+      fundingSource: ModelCredentialFundingSourceSchema,
+      provider: external_exports.string().trim().min(1),
+      // The subscription owner's display name for user_subscription funding, so
+      // the session header can say whose key is being used. Optional for old
+      // writers; null when the name cannot resolve.
+      ownerDisplayName: external_exports.string().trim().min(1).nullable().optional()
+    });
+    SessionUsageCurrentAttributionSchema = external_exports.object({
+      requesterKey: external_exports.string().trim().min(1),
+      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
+      attributionGrade: AttributionGradeSchema
+    }).strip();
+    SessionUsageRequesterSummarySchema = external_exports.object({
+      requesterKey: external_exports.string().trim().min(1).nullable(),
+      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
+      billedCostUsd: NonNegativeUsdSchema2,
+      attributionGrades: external_exports.array(AttributionGradeSchema)
+    }).strip();
+    SessionUsageResponseSchema = external_exports.object({
+      sessionId: SessionIdSchema,
+      totals: UsageTotalsSchema,
+      models: external_exports.array(SessionUsageModelSummarySchema),
+      byFundingSource: external_exports.array(UsageFundingSummarySchema),
+      byRequester: external_exports.array(SessionUsageRequesterSummarySchema).optional().default([]),
+      currentAttribution: SessionUsageCurrentAttributionSchema.nullable().optional().default(null),
+      funding: SessionModelFundingStateSchema.nullable(),
+      firstUsageAt: external_exports.string().datetime().nullable(),
+      lastUsageAt: external_exports.string().datetime().nullable()
+    }).strip();
+    PROJECT_USAGE_RANGE_KEYS = ["7d", "30d"];
+    ProjectUsageRangeKeySchema = external_exports.enum(PROJECT_USAGE_RANGE_KEYS);
+    UtcDateSchema = external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/);
+    ProjectUsageTotalsSchema = UsageTotalsSchema.extend({
+      sessionCount: NonNegativeTokenCountSchema,
+      // Share of all tokens served from prompt-cache reads (0..1); 0 when the
+      // window has no tokens at all.
+      cacheReadShare: external_exports.number().min(0).max(1)
+    });
+    ProjectUsageAgentSummarySchema = UsageTotalsSchema.extend({
+      agentResourceId: external_exports.string().trim().min(1),
+      // Resolved from the agent resource (archived agents included); null only
+      // when the agent row no longer exists at all — the ledger deliberately
+      // outlives its attribution targets.
+      agentName: external_exports.string().trim().min(1).nullable(),
+      sessionCount: NonNegativeTokenCountSchema
+    });
+    ProjectUsageModelSummarySchema = UsageTotalsSchema.extend({
+      model: external_exports.string().trim().min(1)
+    });
+    ProjectUsageSubscriptionSummarySchema = UsageTotalsSchema.extend({
+      userSubscriptionId: UserSubscriptionCredentialIdSchema.nullable(),
+      coveredInferenceCostUsd: NonNegativeUsdSchema2
+    });
+    ProjectUsageSubscriptionSliceSchema = external_exports.object({
+      totals: UsageTotalsSchema.extend({
+        coveredInferenceCostUsd: NonNegativeUsdSchema2
+      }),
+      subscriptions: external_exports.array(ProjectUsageSubscriptionSummarySchema)
+    });
+    ProjectUsageRequesterAttributionSchema = external_exports.object({
+      requesterKey: external_exports.string().trim().min(1).nullable(),
+      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
+      attributionEpochIds: external_exports.array(SessionAttributionEpochIdSchema),
+      attributionGrades: external_exports.array(AttributionGradeSchema)
+    });
+    ProjectUsageRequesterSummarySchema = UsageTotalsSchema.extend({
+      ...ProjectUsageRequesterAttributionSchema.shape,
+      sessionCount: NonNegativeTokenCountSchema
+    });
+    ProjectUsageDayAgentSliceSchema = external_exports.object({
+      agentResourceId: external_exports.string().trim().min(1),
+      billedCostUsd: NonNegativeUsdSchema2
+    });
+    ProjectUsageDayModelSliceSchema = external_exports.object({
+      model: external_exports.string().trim().min(1),
+      billedCostUsd: NonNegativeUsdSchema2
+    });
+    ProjectUsageDayPointSchema = UsageTotalsSchema.extend({
+      date: UtcDateSchema,
+      byAgent: external_exports.array(ProjectUsageDayAgentSliceSchema),
+      byModel: external_exports.array(ProjectUsageDayModelSliceSchema)
+    });
+    ProjectUsageResponseSchema = external_exports.object({
+      organizationId: OrganizationIdSchema,
+      projectId: ProjectIdSchema,
+      range: external_exports.object({
+        key: ProjectUsageRangeKeySchema,
+        since: external_exports.string().datetime(),
+        until: external_exports.string().datetime()
+      }),
+      totals: ProjectUsageTotalsSchema,
+      taskThroughput: external_exports.object({
+        created: external_exports.number().int().nonnegative(),
+        completed: external_exports.number().int().nonnegative(),
+        dropped: external_exports.number().int().nonnegative(),
+        costPerCompletedTaskUsd: NonNegativeUsdSchema2.nullable()
+      }).optional(),
+      byAgent: external_exports.array(ProjectUsageAgentSummarySchema),
+      byModel: external_exports.array(ProjectUsageModelSummarySchema),
+      byFundingSource: external_exports.array(UsageFundingSummarySchema),
+      subscriptionUsage: ProjectUsageSubscriptionSliceSchema,
+      byRequester: external_exports.array(ProjectUsageRequesterSummarySchema),
+      daily: external_exports.array(ProjectUsageDayPointSchema)
+    });
+  }
+});
+
 // ../../packages/schemas/src/session-introspection.ts
-var TruncatedValueSchema, RunConversationSearchSnippetSchema, RunConversationSearchMatchSchema, RunConversationSearchResponseSchema, SessionToolExchangeSchema, SessionToolExchangesResponseSchema, SESSION_TRIGGER_DELIVERY_ACTIONS, SessionTriggerDeliveryActionSchema, SESSION_EVENT_ORIGIN_KINDS, SessionEventOriginKindSchema, SessionTriggerDeliveryRecordSchema, SessionTriggersResponseSchema, RunBindingRecordSchema, RunBindingsResponseSchema, RunTurnSchema, SessionToolStatSchema, RunRuntimeRestartSchema, RunModelFallbackSchema, RunSummarySchema;
+var TruncatedValueSchema, RunConversationSearchSnippetSchema, RunConversationSearchMatchSchema, RunConversationSearchResponseSchema, SessionToolExchangeSchema, SessionToolExchangesResponseSchema, SESSION_TRIGGER_DELIVERY_ACTIONS, SessionTriggerDeliveryActionSchema, SESSION_EVENT_ORIGIN_KINDS, SessionEventOriginKindSchema, SessionTriggerDeliveryRecordSchema, SessionTriggersResponseSchema, RunBindingRecordSchema, RunBindingsResponseSchema, RunTurnSchema, SessionToolStatSchema, RunRuntimeRestartSchema, RunModelFallbackSchema, RUN_USAGE_BREAKDOWN_LIMIT, RUN_USAGE_TURN_LIMIT, RunUsageTotalsSchema, RunUsageHarnessModelSchema, RunUsageTurnSchema, RunUsageSummarySchema, RunSummarySchema;
 var init_session_introspection = __esm({
   "../../packages/schemas/src/session-introspection.ts"() {
     "use strict";
@@ -23017,6 +23248,7 @@ var init_session_introspection = __esm({
     init_runtime_restart_diagnostics();
     init_session_bindings();
     init_sessions();
+    init_usage();
     init_work_provenance();
     TruncatedValueSchema = external_exports.object({
       truncated: external_exports.literal(true),
@@ -23169,6 +23401,62 @@ var init_session_introspection = __esm({
       activatedAt: external_exports.string().datetime().nullable(),
       fallbackModelId: external_exports.string().nullable()
     });
+    RUN_USAGE_BREAKDOWN_LIMIT = 8;
+    RUN_USAGE_TURN_LIMIT = 50;
+    RunUsageTotalsSchema = external_exports.object({
+      eventCount: external_exports.number().int().nonnegative(),
+      inputTokens: external_exports.number().int().nonnegative(),
+      outputTokens: external_exports.number().int().nonnegative(),
+      cacheCreationTokens: external_exports.number().int().nonnegative(),
+      cacheReadTokens: external_exports.number().int().nonnegative(),
+      /**
+       * Stored rate-card-derived model cost; never recomputed by this read.
+       * Omitted when the caller lacks `billing:read`.
+       */
+      derivedCostUsd: external_exports.number().nonnegative().optional(),
+      /**
+       * Sum of provider-reported costs only when every included event reported
+       * one; null for empty, legacy, or partially reported groups, and omitted
+       * when the caller lacks `billing:read`.
+       */
+      providerCostUsd: external_exports.number().nonnegative().nullable().optional()
+    });
+    RunUsageHarnessModelSchema = RunUsageTotalsSchema.extend({
+      harness: UsageHarnessSchema,
+      model: external_exports.string().trim().min(1)
+    });
+    RunUsageTurnSchema = RunUsageTotalsSchema.extend({
+      /** Immutable platform-issued session turn identity. */
+      turnId: SessionCommandIdSchema,
+      /**
+       * Immutable turn creation time. The series is ordered oldest-to-newest by
+       * this field, then turnId, within the bounded most-recent window.
+       */
+      turnCreatedAt: external_exports.string().datetime()
+    });
+    RunUsageSummarySchema = external_exports.object({
+      totals: RunUsageTotalsSchema,
+      /**
+       * Cache reads divided by prompt-side tokens: input + cache creation/write +
+       * cache read. Output tokens are excluded; value is null when the exposed
+       * denominator is zero.
+       */
+      cacheReadRatio: external_exports.object({
+        denominatorTokens: external_exports.number().int().nonnegative(),
+        value: external_exports.number().min(0).max(1).nullable()
+      }),
+      /** Highest derived-cost harness/model groups, in descending spend order. */
+      byHarnessModel: external_exports.array(RunUsageHarnessModelSchema).max(RUN_USAGE_BREAKDOWN_LIMIT),
+      breakdownTruncated: external_exports.boolean(),
+      /** Most recent authoritative turn aggregates, returned oldest-to-newest. */
+      byTurn: external_exports.array(RunUsageTurnSchema).max(RUN_USAGE_TURN_LIMIT),
+      turnBreakdownTruncated: external_exports.boolean(),
+      /**
+       * Current events with no authoritative turn id. These remain in session
+       * totals but are excluded from byTurn rather than guessed into a turn.
+       */
+      unassociated: RunUsageTotalsSchema
+    });
     RunSummarySchema = external_exports.object({
       session: external_exports.object({
         id: SessionIdSchema,
@@ -23284,7 +23572,9 @@ var init_session_introspection = __esm({
         recent: external_exports.array(RunRuntimeRestartSchema)
       }),
       /** Latest same-provider model-fallback activation for the session, if any. */
-      modelFallback: RunModelFallbackSchema
+      modelFallback: RunModelFallbackSchema,
+      /** Current unsuperseded model usage for the session. */
+      usage: RunUsageSummarySchema
     });
   }
 });
@@ -23630,43 +23920,6 @@ var init_runtimes = __esm({
   }
 });
 
-// ../../packages/schemas/src/session-attribution-epochs.ts
-var ATTRIBUTION_GRADES, AttributionGradeSchema, AttributionRequesterSnapshotSchema, SessionAttributionEpochCreateSchema, SessionAttributionEpochRecordSchema;
-var init_session_attribution_epochs = __esm({
-  "../../packages/schemas/src/session-attribution-epochs.ts"() {
-    "use strict";
-    init_zod();
-    init_ids();
-    init_model_credential_funding();
-    init_requester();
-    ATTRIBUTION_GRADES = ["derived", "asserted"];
-    AttributionGradeSchema = external_exports.enum(ATTRIBUTION_GRADES);
-    AttributionRequesterSnapshotSchema = external_exports.object({
-      origin: RequesterRootOriginSchema,
-      userId: UserIdSchema.nullable()
-    }).strict();
-    SessionAttributionEpochCreateSchema = external_exports.object({
-      organizationId: OrganizationIdSchema,
-      projectId: ProjectIdSchema.nullable(),
-      sessionId: SessionIdSchema,
-      sourceCommandId: SessionCommandIdSchema,
-      requesterKey: external_exports.string().trim().min(1),
-      requesterSnapshot: AttributionRequesterSnapshotSchema.nullable(),
-      attributionGrade: AttributionGradeSchema,
-      assertedByCommandId: SessionCommandIdSchema.nullable(),
-      assertedBySessionId: SessionIdSchema.nullable(),
-      assertedByAgentResourceId: external_exports.string().trim().min(1).nullable(),
-      fundingSource: ModelCredentialFundingSourceSchema.nullable(),
-      providerGrantId: ProviderGrantIdSchema.nullable(),
-      userModelSubscriptionId: UserSubscriptionCredentialIdSchema.nullable()
-    }).strict();
-    SessionAttributionEpochRecordSchema = SessionAttributionEpochCreateSchema.extend({
-      id: SessionAttributionEpochIdSchema,
-      createdAt: external_exports.string().datetime()
-    }).strict();
-  }
-});
-
 // ../../packages/schemas/src/spend-cap-runtime.ts
 var SpendCapOverrideStateSchema, SpendCapOverrideUpdateRequestSchema, SPEND_CAP_LIMIT_KINDS, SpendCapLimitKindSchema, EffectiveSpendCapHitSchema, SessionSpendCapRuntimeStatusSchema, AgentSpendCapRuntimeStatusSchema, SessionCreationSpendCapDecisionSchema, SPEND_CAP_RUNTIME_ERROR_CODES, SpendCapRuntimeErrorResponseSchema;
 var init_spend_cap_runtime = __esm({
@@ -57955,6 +58208,23 @@ triggers:
               content: "# Source: https://www.auto.sh/api/v1/templates/%40auto/self-improvement/1.6.0/fragments/environments/agent-runtime.yaml\nharness: claude-code\nenvironment:\n  name: agent-runtime\n  image:\n    kind: preset\n    name: node24\n  resources:\n    memoryMB: 8192\n"
             }
           ]
+        },
+        {
+          version: "1.7.0",
+          files: [
+            {
+              path: "agents/self-improvement-slack.yaml",
+              content: '# Source: https://www.auto.sh/api/v1/templates/%40auto/self-improvement/1.7.0/agents/self-improvement-slack.yaml\n# Required variables: repoFullName, slackChannel, slackConnection\n# Deprecated compatibility entrypoint. New installs should import\n# agents/self-improvement.yaml, whose optional Slack behavior uses the standard\n# `slack` connection and `#dev` channel. This subpath preserves the\n# parameterized, Slack-required, read-only behavior and public agent name of\n# earlier `-slack` versions for existing @latest facades through at least the\n# next minor version.\nimports:\n  - ./self-improvement.yaml\nsystemPrompt: |\n  You are the self-improvement agent for {{ $repoFullName }} and its Auto project.\n  Review real evidence and propose high-leverage improvements to the\n  application or to its Auto agents, prompts, triggers, and processes.\n\n  Evidence sources:\n  - Auto sessions: status, timing, conversations, tool calls, triggers, and\n    transcript search.\n  - GitHub PRs: review comments, expressed preferences, repeated friction,\n    unresolved blockers, and CI failures.\n  - Connected read-only MCP tools: logs, metrics, traces, incidents, support,\n    analytics, and docs. Do not mutate external systems from this workflow.\n\n  Diagnosis standards:\n  - Evidence before verdicts: cite the relevant tool call, event, PR comment,\n    log pattern, or prompt text.\n  - Prefer high-confidence, high-leverage fixes, especially changes the user\n    wants and that can be automated going forward.\n  - A preference need not be repeated before you suggest encoding it; repetition\n    only raises confidence and priority.\n  - Every finding names a concrete app, test, doc, agent, trigger, prompt, or\n    process change.\n  - Your own session\'s past sessions are in scope - scrutinize them like any\n    other run.\n\n  Report format (your final message, every run):\n  1. Verdict - one line: top opportunity, closures, or why more data is needed.\n  2. Findings - each with evidence, affected surface, and the proposed fix.\n  3. Closures - previously reported problems now resolved.\n  4. Deferred - promising leads skipped because they need more evidence.\n\n  Slack protocol ({{ $slackChannel }}): post only when there is something actionable. One\n  short top-level line (sweep time and counts), then exactly one threaded\n  reply with the detail as mrkdwn bullets. Use the threadId returned by\n  chat.send for the reply; never guess thread ids. Links are\n  <https://url|text>.\ninitialPrompt: |\n  A scheduled heartbeat spawned this run (scheduled at\n  "{{heartbeat.scheduledAt}}") to sweep the project\'s recent sessions\n  for failures, anomalies, PR feedback, and improvement opportunities.\n\n  Sweep protocol:\n  - Find your previous report with auto.sessions.list/conversation. Avoid\n    re-reporting old findings; close resolved ones and escalate recurring ones.\n  - Triage recent sessions, PR feedback, and relevant read-only data sources.\n  - Deep-dive at most three evidence clusters. Prefer one well-evidenced,\n    automatable improvement over many shallow observations.\n\n  Deliver per your profile instructions and always end with the four-section\n  report.\n# The Slack variant reports to the channel: pin the github tool list back to\n# the 1.0.0 read-only surface (the base widens it for its tracking issue).\n# Narrow the inherited git mount to read-only too: this variant never\n# writes to GitHub (no tracking issue), so issues drops to read.\nmounts:\n  - mountPath: /workspace/auto\n    auth:\n      capabilities:\n        contents: read\n        pullRequests: read\n        issues: read\n        checks: read\n        actions: read\ntools:\n  github:\n    kind: github\n    tools:\n      - search_pull_requests\n      - pull_request_read\n      - actions_list\n      - actions_get\n  chat:\n    kind: local\n    implementation: chat\n    auth:\n      kind: connection\n      provider: slack\n      connection: "{{ $slackConnection }}"\n      optional: false\ntriggers:\n  - name: mention\n    event: chat.message.mentioned\n    connection: "{{ $slackConnection }}"\n    optional: false\n    where:\n      $.chat.provider: slack\n      $.auto.authored: false\n    message: |\n      {{message.author.userName}} mentioned you on Slack:\n\n      {{message.text}}\n\n      Channel: {{chat.channelId}}\n      Thread: {{chat.threadId}}\n\n      Reply in that thread with chat.send. If the user clearly asks for a\n      sweep, run it. If required context is missing, ask for the time window,\n      target agents, PRs, or data source. Otherwise, briefly explain that you\n      review PR feedback, read-only data sources, and Auto session history,\n      then propose concrete improvements when something is actionable.\n    routing:\n      kind: spawn\n'
+            },
+            {
+              path: "agents/self-improvement.yaml",
+              content: '# Source: https://www.auto.sh/api/v1/templates/%40auto/self-improvement/1.7.0/agents/self-improvement.yaml\n# Required variables: repoFullName\nname: self-improvement\nmodel:\n  provider: anthropic\n  id: claude-opus-4-8\nidentity:\n  displayName: Self Improvement\n  username: self-improvement\n  avatar:\n    asset: .auto/assets/self-improvement.png\n    sha256: 5f8e96bb0919d0fc689e1593b70a2b0c2c28913c210c76b7e2d3d5f22a94b1dd\n  description: Reviews PR feedback, read-only data, and Auto sessions to propose concrete improvements, with optional Slack delivery.\nimports:\n  - ../fragments/environments/agent-runtime.yaml\nbindings:\n  slack.thread:\n    continuity: agent\nsystemPrompt: |\n  You are the self-improvement agent for {{ $repoFullName }} and its Auto project.\n  Review real evidence and propose high-leverage improvements to the\n  application or to its Auto agents, prompts, triggers, tools, and processes.\n  Treat the repository\'s committed `.auto/` files as first-class evidence:\n  inspect the actual agent resources when proposing a change to the factory,\n  and point to the exact file and field that should change.\n\n  Evidence sources:\n  - Auto sessions: status, timing, conversations, tool calls, triggers, and\n    transcript search.\n  - GitHub PRs: review comments, expressed preferences, repeated friction,\n    unresolved blockers, and CI failures.\n  - Connected read-only MCP tools: logs, metrics, traces, incidents, support,\n    analytics, and docs. Do not mutate external systems from this workflow.\n\n  Diagnosis standards:\n  - Evidence before verdicts: cite the relevant tool call, event, PR comment,\n    log pattern, or prompt text.\n  - Prefer high-confidence, high-leverage fixes, especially changes the user\n    wants and that can be automated going forward.\n  - A preference need not be repeated before you suggest encoding it; repetition\n    only raises confidence and priority.\n  - Every finding names a concrete app, test, doc, agent, trigger, prompt, or\n    process change.\n  - Your own session\'s past sessions are in scope - scrutinize them like any\n    other run.\n\n  Report format (your final message, every run):\n  1. Verdict - one line: top opportunity, closures, or why more data is needed.\n  2. Findings - each with evidence, affected surface, and the proposed fix.\n  3. Closures - previously reported problems now resolved.\n  4. Deferred - promising leads skipped because they need more evidence.\n\n  Delivery: the report is this run\'s final message. Compare against\n  earlier sweeps by reading your previous sessions with\n  auto.sessions.list/conversation - the session history is the report\n  history. Sweep findings routinely cite session internals and PR\n  friction, so they do not belong on GitHub by default.\n\n  Fallback (only when the team has explicitly asked for tracking-issue\n  delivery, and never on a public repository): keep a single tracking\n  issue titled "Self-improvement sweep reports" - find it with\n  search_issues, create it with issue_write only if it is missing, and\n  add exactly one comment per sweep with add_issue_comment: one short\n  first line (sweep time and counts), then the detail as Markdown\n  bullets. Never open a new issue per finding, and post only when there\n  is something actionable.\n\n  Slack delivery is optional and uses the standard `slack` connection and\n  `#dev` channel. When the chat tool is available, post only when there is\n  something actionable: one short top-level line with the sweep time and\n  counts, then exactly one threaded reply with the detail as mrkdwn bullets.\n  Use the threadId returned by chat.send for the reply; never guess thread ids.\n  When the tool is unavailable, do not treat Slack delivery as a failure; the\n  run report remains the complete sweep. If a user asks for Slack delivery\n  while it is unavailable, offer to connect the standard `slack` connection\n  and explain that a fresh apply and session make the capability available.\n\n  Scheduled sweep closure:\n  - Finish the final four-section report/reply and, for actionable findings,\n    every required Slack or Chief handoff. Only after every owed delivery is\n    complete, call `auto.sessions.complete_current` exactly once with a compact\n    one-line handoff.\n  - This applies to actionable and no-action sweeps. A no-findings sweep with\n    Verdict "Nothing actionable." completes after its final four-section report;\n    it owes no Slack or Chief handoff.\n  - Never call `auto.sessions.archive_current` to close a sweep. Do not call\n    `auto.unbind`, `auto.chat.unsubscribe`, or otherwise release a Slack\n    thread binding during closure. The configured agent-continuity\n    `slack.thread` binding survives completion while lifecycle remains manual.\n    `chat_send` stamps that continuity from this spec. It routes a human reply\n    back to this completed holder and reopens this session.\ninitialPrompt: |\n  A scheduled heartbeat spawned this run (scheduled at\n  "{{heartbeat.scheduledAt}}") to sweep the project\'s recent sessions\n  for failures, anomalies, PR feedback, and improvement opportunities.\n\n  Sweep protocol:\n  - Find your previous report with auto.sessions.list/conversation. Avoid\n    re-reporting old findings; close resolved ones and escalate recurring ones.\n  - Triage recent sessions, PR feedback, and relevant read-only data sources.\n  - Deep-dive at most three evidence clusters. Prefer one well-evidenced,\n    automatable improvement over many shallow observations.\n\n  Deliver per your profile instructions and always end with the four-section\n  report.\nmounts:\n  # GitHub App git mount provisions the GitHub MCP proxy (the github\n  # tools below are broken without a githubApp mount) and stages a\n  # read-only checkout the sweep can ground findings in. Capabilities\n  # line up with the declared tools: pullRequests/issues read for PR\n  # comment and tracking-issue inspection, issues: write for the opt-in\n  # tracking-issue fallback (issue_write/add_issue_comment), actions/checks\n  # read for CI. No merge/secrets/workflows.\n  - kind: git\n    repository: "{{ $repoFullName }}"\n    mountPath: /workspace/auto\n    ref: main\n    auth:\n      kind: githubApp\n      capabilities:\n        contents: read\n        pullRequests: read\n        issues: write\n        checks: read\n        actions: read\ntools:\n  auto:\n    kind: local\n    implementation: auto\n  github:\n    kind: github\n    tools:\n      - search_pull_requests\n      - pull_request_read\n      - actions_list\n      - actions_get\n      - search_issues\n      - issue_read\n      - issue_write\n      - add_issue_comment\n  chat:\n    kind: local\n    implementation: chat\n    auth:\n      kind: connection\n      provider: slack\n      connection: slack\n      optional: true\ntriggers:\n  - name: sweep-heartbeat\n    kind: heartbeat\n    cron: 0 */2 * * *\n    timezone: UTC\n    routing:\n      kind: spawn\n  - name: mention\n    event: chat.message.mentioned\n    connection: slack\n    optional: true\n    where:\n      $.chat.provider: slack\n      $.auto.authored: false\n    message: |\n      {{message.author.userName}} mentioned you on Slack:\n\n      {{message.text}}\n\n      Channel: {{chat.channelId}}\n      Thread: {{chat.threadId}}\n\n      Reply in that thread with chat.send. If the user clearly asks for a\n      sweep, run it. If required context is missing, ask for the time window,\n      target agents, PRs, or data source. Otherwise, briefly explain that you\n      review PR feedback, read-only data sources, and Auto session history,\n      then propose concrete improvements when something is actionable.\n    routing:\n      kind: spawn\n'
+            },
+            {
+              path: "fragments/environments/agent-runtime.yaml",
+              content: "# Source: https://www.auto.sh/api/v1/templates/%40auto/self-improvement/1.7.0/fragments/environments/agent-runtime.yaml\nharness: claude-code\nenvironment:\n  name: agent-runtime\n  image:\n    kind: preset\n    name: node24\n  resources:\n    memoryMB: 8192\n"
+            }
+          ]
         }
       ],
       "@auto/slopbusters": [
@@ -81718,195 +81988,6 @@ var init_temporal = __esm({
   }
 });
 
-// ../../packages/schemas/src/usage.ts
-var UsageHarnessSchema, NonNegativeTokenCountSchema, NonNegativeUsdSchema2, UsageEventSchema, UsageEventRecordSchema, UsageTotalsSchema, SessionUsageModelSummarySchema, UsageFundingSummarySchema, SessionModelFundingStateSchema, SessionUsageCurrentAttributionSchema, SessionUsageRequesterSummarySchema, SessionUsageResponseSchema, PROJECT_USAGE_RANGE_KEYS, ProjectUsageRangeKeySchema, UtcDateSchema, ProjectUsageTotalsSchema, ProjectUsageAgentSummarySchema, ProjectUsageModelSummarySchema, ProjectUsageSubscriptionSummarySchema, ProjectUsageSubscriptionSliceSchema, ProjectUsageRequesterAttributionSchema, ProjectUsageRequesterSummarySchema, ProjectUsageDayAgentSliceSchema, ProjectUsageDayModelSliceSchema, ProjectUsageDayPointSchema, ProjectUsageResponseSchema;
-var init_usage = __esm({
-  "../../packages/schemas/src/usage.ts"() {
-    "use strict";
-    init_zod();
-    init_agents();
-    init_ids();
-    init_model_credential_funding();
-    init_session_attribution_epochs();
-    init_model_credential_funding();
-    UsageHarnessSchema = external_exports.enum(AGENT_HARNESSES);
-    NonNegativeTokenCountSchema = external_exports.number().int().nonnegative();
-    NonNegativeUsdSchema2 = external_exports.number().nonnegative();
-    UsageEventSchema = external_exports.object({
-      // Tenant + lineage attribution. projectId is nullable because a resource can
-      // belong directly to an organization with no project.
-      organizationId: OrganizationIdSchema,
-      projectId: ProjectIdSchema.nullable(),
-      sessionId: SessionIdSchema,
-      agentResourceId: external_exports.string().trim().min(1),
-      harness: UsageHarnessSchema,
-      // The concrete model id the call billed against (e.g. "claude-opus-4-8"),
-      // taken from the response rather than any selection layer — a single run can
-      // touch multiple models, so this is per-call, never per-run.
-      model: external_exports.string().trim().min(1),
-      inputTokens: NonNegativeTokenCountSchema,
-      outputTokens: NonNegativeTokenCountSchema,
-      cacheCreationTokens: NonNegativeTokenCountSchema,
-      cacheReadTokens: NonNegativeTokenCountSchema,
-      // The provider's own cost when it reports one, kept only for reconciliation.
-      // Nullable/absent because it can be 0 or missing — never the billing source.
-      providerCostUsd: NonNegativeUsdSchema2.nullable(),
-      // Platform-derived cost and the rate-card version it was derived under, so a
-      // later rate-card change never silently rewrites historical cost.
-      derivedCostUsd: NonNegativeUsdSchema2,
-      pricingVersion: external_exports.string().trim().min(1),
-      // What this call debits from the organization's credit balance, and the fee
-      // card it was derived under (see `./billing`). Required — a writer that
-      // forgets it would silently record unbilled usage, so there is no default.
-      billedCostUsd: NonNegativeUsdSchema2,
-      billingFeeVersion: external_exports.string().trim().min(1),
-      // Historical funding attribution for the model call. The provider grant id is
-      // denormalized and nullable for platform-funded usage, matching the ledger's
-      // no-FK audit-trail contract.
-      fundingSource: ModelCredentialFundingSourceSchema.default("platform"),
-      modelCredentialProviderGrantId: ProviderGrantIdSchema.nullable().default(null),
-      // Optional during the cross-chunk rollout: B1 defines the attribution field
-      // before B3 deploys its persistence column and D3 begins populating it.
-      modelCredentialUserSubscriptionId: UserSubscriptionCredentialIdSchema.nullable().optional(),
-      attributionEpochId: SessionAttributionEpochIdSchema.nullable().optional(),
-      requesterKey: external_exports.string().trim().min(1).nullable().optional(),
-      attributionGrade: AttributionGradeSchema.nullable().optional(),
-      occurredAt: external_exports.string().datetime()
-    });
-    UsageEventRecordSchema = UsageEventSchema.extend({
-      id: external_exports.string().trim().min(1),
-      attributionEpochId: SessionAttributionEpochIdSchema.nullable().default(null),
-      requesterKey: external_exports.string().trim().min(1).nullable().default(null),
-      attributionGrade: AttributionGradeSchema.nullable().default(null),
-      // Set on append-only correction rows: this row replaces the pointed-at row's
-      // amounts, and ledger readers exclude any row another row supersedes.
-      supersedesUsageEventId: external_exports.string().trim().min(1).nullable().default(null),
-      createdAt: external_exports.string().datetime()
-    });
-    UsageTotalsSchema = external_exports.object({
-      eventCount: NonNegativeTokenCountSchema,
-      inputTokens: NonNegativeTokenCountSchema,
-      outputTokens: NonNegativeTokenCountSchema,
-      cacheCreationTokens: NonNegativeTokenCountSchema,
-      cacheReadTokens: NonNegativeTokenCountSchema,
-      totalTokens: NonNegativeTokenCountSchema,
-      billedCostUsd: NonNegativeUsdSchema2
-    });
-    SessionUsageModelSummarySchema = UsageTotalsSchema.extend({
-      model: external_exports.string().trim().min(1)
-    });
-    UsageFundingSummarySchema = UsageTotalsSchema.extend({
-      fundingSource: ModelCredentialFundingSourceSchema,
-      coveredInferenceCostUsd: NonNegativeUsdSchema2
-    });
-    SessionModelFundingStateSchema = external_exports.object({
-      fundingSource: ModelCredentialFundingSourceSchema,
-      provider: external_exports.string().trim().min(1),
-      // The subscription owner's display name for user_subscription funding, so
-      // the session header can say whose key is being used. Optional for old
-      // writers; null when the name cannot resolve.
-      ownerDisplayName: external_exports.string().trim().min(1).nullable().optional()
-    });
-    SessionUsageCurrentAttributionSchema = external_exports.object({
-      requesterKey: external_exports.string().trim().min(1),
-      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
-      attributionGrade: AttributionGradeSchema
-    }).strip();
-    SessionUsageRequesterSummarySchema = external_exports.object({
-      requesterKey: external_exports.string().trim().min(1).nullable(),
-      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
-      billedCostUsd: NonNegativeUsdSchema2,
-      attributionGrades: external_exports.array(AttributionGradeSchema)
-    }).strip();
-    SessionUsageResponseSchema = external_exports.object({
-      sessionId: SessionIdSchema,
-      totals: UsageTotalsSchema,
-      models: external_exports.array(SessionUsageModelSummarySchema),
-      byFundingSource: external_exports.array(UsageFundingSummarySchema),
-      byRequester: external_exports.array(SessionUsageRequesterSummarySchema).optional().default([]),
-      currentAttribution: SessionUsageCurrentAttributionSchema.nullable().optional().default(null),
-      funding: SessionModelFundingStateSchema.nullable(),
-      firstUsageAt: external_exports.string().datetime().nullable(),
-      lastUsageAt: external_exports.string().datetime().nullable()
-    }).strip();
-    PROJECT_USAGE_RANGE_KEYS = ["7d", "30d"];
-    ProjectUsageRangeKeySchema = external_exports.enum(PROJECT_USAGE_RANGE_KEYS);
-    UtcDateSchema = external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/);
-    ProjectUsageTotalsSchema = UsageTotalsSchema.extend({
-      sessionCount: NonNegativeTokenCountSchema,
-      // Share of all tokens served from prompt-cache reads (0..1); 0 when the
-      // window has no tokens at all.
-      cacheReadShare: external_exports.number().min(0).max(1)
-    });
-    ProjectUsageAgentSummarySchema = UsageTotalsSchema.extend({
-      agentResourceId: external_exports.string().trim().min(1),
-      // Resolved from the agent resource (archived agents included); null only
-      // when the agent row no longer exists at all — the ledger deliberately
-      // outlives its attribution targets.
-      agentName: external_exports.string().trim().min(1).nullable(),
-      sessionCount: NonNegativeTokenCountSchema
-    });
-    ProjectUsageModelSummarySchema = UsageTotalsSchema.extend({
-      model: external_exports.string().trim().min(1)
-    });
-    ProjectUsageSubscriptionSummarySchema = UsageTotalsSchema.extend({
-      userSubscriptionId: UserSubscriptionCredentialIdSchema.nullable(),
-      coveredInferenceCostUsd: NonNegativeUsdSchema2
-    });
-    ProjectUsageSubscriptionSliceSchema = external_exports.object({
-      totals: UsageTotalsSchema.extend({
-        coveredInferenceCostUsd: NonNegativeUsdSchema2
-      }),
-      subscriptions: external_exports.array(ProjectUsageSubscriptionSummarySchema)
-    });
-    ProjectUsageRequesterAttributionSchema = external_exports.object({
-      requesterKey: external_exports.string().trim().min(1).nullable(),
-      requesterDisplayName: external_exports.string().trim().min(1).nullable(),
-      attributionEpochIds: external_exports.array(SessionAttributionEpochIdSchema),
-      attributionGrades: external_exports.array(AttributionGradeSchema)
-    });
-    ProjectUsageRequesterSummarySchema = UsageTotalsSchema.extend({
-      ...ProjectUsageRequesterAttributionSchema.shape,
-      sessionCount: NonNegativeTokenCountSchema
-    });
-    ProjectUsageDayAgentSliceSchema = external_exports.object({
-      agentResourceId: external_exports.string().trim().min(1),
-      billedCostUsd: NonNegativeUsdSchema2
-    });
-    ProjectUsageDayModelSliceSchema = external_exports.object({
-      model: external_exports.string().trim().min(1),
-      billedCostUsd: NonNegativeUsdSchema2
-    });
-    ProjectUsageDayPointSchema = UsageTotalsSchema.extend({
-      date: UtcDateSchema,
-      byAgent: external_exports.array(ProjectUsageDayAgentSliceSchema),
-      byModel: external_exports.array(ProjectUsageDayModelSliceSchema)
-    });
-    ProjectUsageResponseSchema = external_exports.object({
-      organizationId: OrganizationIdSchema,
-      projectId: ProjectIdSchema,
-      range: external_exports.object({
-        key: ProjectUsageRangeKeySchema,
-        since: external_exports.string().datetime(),
-        until: external_exports.string().datetime()
-      }),
-      totals: ProjectUsageTotalsSchema,
-      taskThroughput: external_exports.object({
-        created: external_exports.number().int().nonnegative(),
-        completed: external_exports.number().int().nonnegative(),
-        dropped: external_exports.number().int().nonnegative(),
-        costPerCompletedTaskUsd: NonNegativeUsdSchema2.nullable()
-      }).optional(),
-      byAgent: external_exports.array(ProjectUsageAgentSummarySchema),
-      byModel: external_exports.array(ProjectUsageModelSummarySchema),
-      byFundingSource: external_exports.array(UsageFundingSummarySchema),
-      subscriptionUsage: ProjectUsageSubscriptionSliceSchema,
-      byRequester: external_exports.array(ProjectUsageRequesterSummarySchema),
-      daily: external_exports.array(ProjectUsageDayPointSchema)
-    });
-  }
-});
-
 // ../../packages/schemas/src/user-subscriptions.ts
 var ATTRIBUTION_COLD_BOUNDARY_IDLE_MS, USER_SUBSCRIPTION_PROVIDERS, UserSubscriptionProviderSchema, USER_SUBSCRIPTION_CREDENTIAL_STATUSES, UserSubscriptionCredentialStatusSchema, OpenAiUserSubscriptionCredentialSecretSchema, UserSubscriptionCredentialSchema, USER_SUBSCRIPTION_POLICY_MODES, UserSubscriptionPolicyModeSchema, USER_SUBSCRIPTION_POLICY_ORIGINS, UserSubscriptionPolicyOriginSchema, UserSubscriptionPolicySchema, UserSubscriptionFundingCandidateSchema, UserSubscriptionFundingDecisionInputSchema, USER_SUBSCRIPTION_REJECTION_REASONS, UserSubscriptionRejectionReasonSchema, USER_SUBSCRIPTION_SELECTION_REASONS, UserSubscriptionSelectionReasonSchema, UserSubscriptionFundingDecisionSchema;
 var init_user_subscriptions = __esm({
@@ -84895,7 +84976,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "@autohq/cli",
-      version: "0.1.564",
+      version: "0.1.566",
       license: "SEE LICENSE IN README.md",
       publishConfig: {
         access: "public"
@@ -96097,13 +96178,23 @@ var RuntimeBridgeOutputLivenessEnvelopeSchema = external_exports.object({
   intervalMs: external_exports.number().int().positive(),
   createdAt: external_exports.string().datetime()
 });
+var RuntimeBridgeOutputUsageTurnEnvelopeSchema = external_exports.object({
+  type: external_exports.literal("runtime.usage_turn"),
+  sessionId: SessionIdSchema2,
+  runtimeId: RuntimeIdSchema2,
+  bridgeLeaseId: RuntimeBridgeLeaseIdSchema2,
+  outputSeq: external_exports.number().int().positive(),
+  turnId: SessionCommandIdSchema2.nullable(),
+  createdAt: external_exports.string().datetime()
+});
 var RuntimeBridgeOutputEnvelopeSchema = external_exports.union([
   RuntimeBridgeOutputEntryEnvelopeSchema,
   RuntimeBridgeOutputDeltaEnvelopeSchema,
   RuntimeBridgeOutputUiMessageChunkEnvelopeSchema,
   RuntimeBridgeOutputUiMessagePartEnvelopeSchema,
   RuntimeBridgeOutputUiMessageCompletedEnvelopeSchema,
-  RuntimeBridgeOutputLivenessEnvelopeSchema
+  RuntimeBridgeOutputLivenessEnvelopeSchema,
+  RuntimeBridgeOutputUsageTurnEnvelopeSchema
 ]);
 function isLiveOnlyRuntimeOutputEnvelope(envelope) {
   return envelope.type === "ui.message.chunk" || envelope.type === "conversation.delta";
@@ -97750,6 +97841,9 @@ function replacePayloadWithMarker(envelope, originalBytes) {
           parts: [{ type: "text", text, state: "done" }]
         }
       };
+    case "runtime.liveness":
+    case "runtime.usage_turn":
+      return envelope;
   }
 }
 
@@ -98183,6 +98277,23 @@ var AgentBridgeOutputBuffer = class {
     );
     await this.drainPendingOutputs();
   }
+  /**
+   * Publish and await the active immutable platform turn before model work.
+   * Unlike liveness, every transition is retained and replayed: skipping one
+   * could attribute a billed call to the preceding turn.
+   */
+  async emitUsageTurn(context, turnId) {
+    this.outputSeq += 1;
+    this.enqueueOutput(
+      buildUsageTurnOutputEnvelope({
+        context,
+        outputSeq: this.outputSeq,
+        turnId,
+        createdAt: now2()
+      })
+    );
+    await this.drainPendingOutputs({ failIfBlocked: true });
+  }
   async emitUiMessageChunk(context, projection) {
     await this.flushPendingDelta();
     await this.emitLiveUiMessageChunk(context, projection.chunk);
@@ -98963,6 +99074,18 @@ function buildLivenessOutputEnvelope(input) {
     createdAt: input.createdAt
   });
 }
+function buildUsageTurnOutputEnvelope(input) {
+  const { context } = input;
+  return RuntimeBridgeOutputEnvelopeSchema.parse({
+    type: "runtime.usage_turn",
+    sessionId: context.sessionId,
+    runtimeId: context.runtimeId,
+    bridgeLeaseId: context.bridgeLeaseId,
+    outputSeq: input.outputSeq,
+    turnId: input.turnId,
+    createdAt: input.createdAt
+  });
+}
 function buildUiMessagePartOutputEnvelope(input) {
   const { context, snapshot } = input;
   return RuntimeBridgeOutputEnvelopeSchema.parse({
@@ -100254,7 +100377,7 @@ var ClaudeAgentBridgeSessionImpl = class {
       return;
     }
     await this.awaitMcpRegistration();
-    this.enqueueUserMessage(message, commandIds);
+    await this.enqueueUserMessage(message, commandIds);
     this.input.runtimeLogger?.info("agent_bridge_claude_send_message_queued", {
       delivery_mode: mode,
       duration_ms: Date.now() - startedAt,
@@ -100360,6 +100483,7 @@ var ClaudeAgentBridgeSessionImpl = class {
     }
     this.openTurnCommandIds.length = 0;
     this.publishActiveTurnCommandIds();
+    void this.input.onUsageTurnChanged?.(null);
     this.stderr.flush();
     this.inputQueue.close();
     if (previousState.kind === "started") {
@@ -100453,6 +100577,7 @@ var ClaudeAgentBridgeSessionImpl = class {
               consumedCommandIds: this.openTurnCommandIds.shift() ?? []
             };
             this.publishActiveTurnCommandIds();
+            await this.input.onUsageTurnChanged?.(null);
             this.pendingToolUses.clear();
             this.interruptSettling = false;
             this.turnResultCount += 1;
@@ -100464,7 +100589,7 @@ var ClaudeAgentBridgeSessionImpl = class {
               consumed_command_ids: meta3.consumedCommandIds
             });
             if (this.activeTurnCount === 0) {
-              this.flushDeferredMessages();
+              await this.flushDeferredMessages();
             }
             if (this.deferredMessages.length === 0) {
               this.cancelledToolRuns = [];
@@ -100494,6 +100619,7 @@ var ClaudeAgentBridgeSessionImpl = class {
         this.activeTurnCount = 0;
         this.openTurnCommandIds.length = 0;
         this.publishActiveTurnCommandIds();
+        await this.input.onUsageTurnChanged?.(null);
         this.pendingToolUses.clear();
         this.interruptSettling = false;
         this.disarmToolBoundaryWait();
@@ -100506,7 +100632,8 @@ var ClaudeAgentBridgeSessionImpl = class {
   // lets injection work while the SDK session is still starting; the queue
   // drains once the query attaches. Startup failures surface through onError
   // instead of failing injection.
-  enqueueUserMessage(message, commandIds) {
+  async enqueueUserMessage(message, commandIds) {
+    await this.input.onUsageTurnChanged?.(commandIds[0] ?? null);
     this.inputQueue.push(claudeAgentUserMessage(message));
     this.activeTurnCount += 1;
     this.openTurnCommandIds.push(commandIds);
@@ -100614,7 +100741,7 @@ var ClaudeAgentBridgeSessionImpl = class {
     clearTimeout(this.toolBoundaryWait.timer);
     this.toolBoundaryWait = null;
   }
-  flushDeferredMessages() {
+  async flushDeferredMessages() {
     if (this.deferredMessages.length === 0) {
       return;
     }
@@ -100625,7 +100752,7 @@ var ClaudeAgentBridgeSessionImpl = class {
       `agent_bridge_claude_deferred_flush count=${pending.length} cancelled_tool_count=${cancelledToolRuns.length}`
     );
     const text = pending.map((entry) => entry.text).join("\n\n");
-    this.enqueueUserMessage(
+    await this.enqueueUserMessage(
       cancelledToolRuns.length > 0 ? `${text}
 
 ${cancelledToolPlatformNotice(cancelledToolRuns)}` : text,
@@ -101542,6 +101669,13 @@ var ClaudeCodeCommandHandler = class {
       RUNTIME_LIVENESS_INTERVAL_MS
     );
   }
+  async emitUsageTurn(turnId) {
+    const activeContext = this.context;
+    if (!activeContext) {
+      throw new Error("Cannot publish usage turn without bridge context");
+    }
+    await this.outputBuffer.emitUsageTurn(activeContext, turnId);
+  }
   async emitBridgeOutput(activeContext, projection) {
     try {
       await this.outputBuffer.emitProjection(activeContext, projection);
@@ -101709,6 +101843,7 @@ var ClaudeCodeCommandHandler = class {
         }
       },
       onActiveTurnCommandIds: (commandIds) => this.autoMcpShim.setActiveTurnCommandIds(commandIds),
+      onUsageTurnChanged: (turnId) => this.emitUsageTurn(turnId),
       canUseTool: this.canUseTool,
       onMessage: (message, meta3) => this.handleAgentMessage(message, meta3),
       onError: (error51) => this.handleAgentError(error51),
@@ -102922,6 +103057,7 @@ var CodexAgentBridgeSessionImpl = class {
     this.pendingCommandIds = [...commandIds];
     this.publishActiveTurnCommandIds(commandIds);
     try {
+      await this.input.onUsageTurnChanged?.(commandIds[0] ?? null);
       await this.request("turn/start", {
         threadId,
         input: [userTextInput(message)]
@@ -102929,6 +103065,7 @@ var CodexAgentBridgeSessionImpl = class {
     } catch (error51) {
       this.pendingCommandIds = [];
       this.publishActiveTurnCommandIds();
+      await this.input.onUsageTurnChanged?.(null);
       throw error51;
     }
   }
@@ -102936,6 +103073,7 @@ var CodexAgentBridgeSessionImpl = class {
     this.pendingCommandIds = [...commandIds];
     this.publishActiveTurnCommandIds(commandIds);
     try {
+      await this.input.onUsageTurnChanged?.(commandIds[0] ?? null);
       await this.request("turn/start", {
         threadId,
         input: []
@@ -102943,6 +103081,7 @@ var CodexAgentBridgeSessionImpl = class {
     } catch (error51) {
       this.pendingCommandIds = [];
       this.publishActiveTurnCommandIds();
+      await this.input.onUsageTurnChanged?.(null);
       throw error51;
     }
   }
@@ -103306,6 +103445,7 @@ var CodexAgentBridgeSessionImpl = class {
       this.activeTurnId = null;
       this.activeCommandIds = [];
       this.publishActiveTurnCommandIds();
+      void this.input.onUsageTurnChanged?.(null);
     }
     this.pendingToolItemIds.clear();
     this.resolveSettlement();
@@ -103321,6 +103461,7 @@ var CodexAgentBridgeSessionImpl = class {
     this.activeCommandIds = [];
     this.pendingCommandIds = [];
     this.publishActiveTurnCommandIds();
+    void this.input.onUsageTurnChanged?.(null);
     this.pendingToolItemIds.clear();
     this.resolveSettlement();
     this.input.writeOutput?.(`agent_bridge_codex_exited code=${code ?? ""}`);
@@ -103716,6 +103857,13 @@ var CodexCommandHandler = class {
       RUNTIME_LIVENESS_INTERVAL_MS
     );
   }
+  async emitUsageTurn(turnId) {
+    const activeContext = this.context;
+    if (!activeContext) {
+      throw new Error("Cannot publish usage turn without bridge context");
+    }
+    await this.outputBuffer.emitUsageTurn(activeContext, turnId);
+  }
   async emit(activeContext, projection) {
     try {
       await this.outputBuffer.emitProjection(activeContext, projection);
@@ -103738,6 +103886,7 @@ var CodexCommandHandler = class {
       onServerRequest: (request) => this.handleServerRequest(request),
       onThreadId: (threadId) => this.persistThreadId(threadId),
       onActiveTurnCommandIds: (commandIds) => this.autoMcpShim.setActiveTurnCommandIds(commandIds),
+      onUsageTurnChanged: (turnId) => this.emitUsageTurn(turnId),
       onError: (error51) => this.handleSessionError(error51),
       onExit: () => {
         if (this.session === session) {