@jinn-network/client 0.1.7 → 0.1.8-canary.09a3b2f6

package/dist/spend/usage.js

@@ -0,0 +1,113 @@
+import { readFileSync } from 'node:fs';
+import { join } from 'node:path';
+import { CLAUDE_CODE_HARNESS, CODEX_HARNESS, canonicalHarnessName } from '../harnesses/names.js';
+import { estimateModelCost } from '../harnesses/cost-estimates.js';
+import { priceTokens } from './pricing.js';
+/** USD attributed to a task whose model has no known price. */
+export const UNKNOWN_MODEL_FALLBACK_USD = 1.0;
+/** Parse Claude Code `--output-format stream-json` output for the terminal result. */
+export function parseClaudeCodeUsage(stdoutJsonl) {
+    let result = null;
+    for (const line of stdoutJsonl.split('\n')) {
+        const trimmed = line.trim();
+        if (!trimmed)
+            continue;
+        let obj;
+        try {
+            obj = JSON.parse(trimmed);
+        }
+        catch {
+            continue;
+        }
+        if (obj['type'] === 'result' && typeof obj['total_cost_usd'] === 'number') {
+            const usage = obj['usage'];
+            result = {
+                costUsd: obj['total_cost_usd'],
+                inputTokens: typeof usage?.['input_tokens'] === 'number' ? usage['input_tokens'] : undefined,
+                outputTokens: typeof usage?.['output_tokens'] === 'number' ? usage['output_tokens'] : undefined,
+            };
+        }
+    }
+    return result;
+}
+/** Parse Codex `--json` output for the last turn.completed token usage. */
+export function parseCodexUsage(stdoutJsonl) {
+    let result = null;
+    for (const line of stdoutJsonl.split('\n')) {
+        const trimmed = line.trim();
+        if (!trimmed)
+            continue;
+        let obj;
+        try {
+            obj = JSON.parse(trimmed);
+        }
+        catch {
+            continue;
+        }
+        if (obj['type'] === 'turn.completed' && obj['usage']) {
+            const usage = obj['usage'];
+            const inT = usage['input_tokens'];
+            const outT = usage['output_tokens'];
+            if (typeof inT === 'number' && typeof outT === 'number') {
+                result = { inputTokens: inT, outputTokens: outT };
+            }
+        }
+    }
+    return result;
+}
+function heuristicUsage(model) {
+    const est = model ? estimateModelCost(model) : null;
+    return {
+        model: model ?? 'unknown',
+        costUsd: est?.usd ?? UNKNOWN_MODEL_FALLBACK_USD,
+        estimated: true,
+    };
+}
+/**
+ * Determine the USD cost of a finished harness run. Reads the harness's own
+ * output file for observed usage; falls back to a heuristic on any failure.
+ * Always returns a HarnessUsage — never throws.
+ */
+export function harvestHarnessUsage(harness, workingDir, model) {
+    try {
+        const canonical = canonicalHarnessName(harness);
+        if (canonical === CLAUDE_CODE_HARNESS) {
+            const raw = readFileSync(join(workingDir, '.claude-code', 'stdout.jsonl'), 'utf8');
+            const parsed = parseClaudeCodeUsage(raw);
+            if (parsed) {
+                return {
+                    model: model ?? 'unknown',
+                    costUsd: parsed.costUsd,
+                    estimated: false,
+                    inputTokens: parsed.inputTokens,
+                    outputTokens: parsed.outputTokens,
+                };
+            }
+            return heuristicUsage(model);
+        }
+        if (canonical === CODEX_HARNESS) {
+            const raw = readFileSync(join(workingDir, '.codex-code', 'stdout.jsonl'), 'utf8');
+            const parsed = parseCodexUsage(raw);
+            // When there is no model id, parsed token counts cannot be priced —
+            // discard them and fall back to the heuristic (per spec).
+            if (parsed && model) {
+                const usd = priceTokens(model, parsed);
+                if (usd != null) {
+                    return {
+                        model,
+                        costUsd: usd,
+                        estimated: false,
+                        inputTokens: parsed.inputTokens,
+                        outputTokens: parsed.outputTokens,
+                    };
+                }
+            }
+            return heuristicUsage(model);
+        }
+        return heuristicUsage(model);
+    }
+    catch {
+        return heuristicUsage(model);
+    }
+}
+//# sourceMappingURL=usage.js.map

package/dist/spend/usage.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"usage.js","sourceRoot":"","sources":["../../src/spend/usage.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,mBAAmB,EAAE,aAAa,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AACjG,OAAO,EAAE,iBAAiB,EAAE,MAAM,gCAAgC,CAAC;AACnE,OAAO,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAE3C,+DAA+D;AAC/D,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAW9C,sFAAsF;AACtF,MAAM,UAAU,oBAAoB,CAClC,WAAmB;IAEnB,IAAI,MAAM,GAA4E,IAAI,CAAC;IAC3F,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,IAAI,CAAC,OAAO;YAAE,SAAS;QACvB,IAAI,GAA4B,CAAC;QACjC,IAAI,CAAC;YAAC,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC;YAAC,SAAS;QAAC,CAAC;QACtD,IAAI,GAAG,CAAC,MAAM,CAAC,KAAK,QAAQ,IAAI,OAAO,GAAG,CAAC,gBAAgB,CAAC,KAAK,QAAQ,EAAE,CAAC;YAC1E,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAAwC,CAAC;YAClE,MAAM,GAAG;gBACP,OAAO,EAAE,GAAG,CAAC,gBAAgB,CAAW;gBACxC,WAAW,EAAE,OAAO,KAAK,EAAE,CAAC,cAAc,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,cAAc,CAAW,CAAC,CAAC,CAAC,SAAS;gBACtG,YAAY,EAAE,OAAO,KAAK,EAAE,CAAC,eAAe,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAW,CAAC,CAAC,CAAC,SAAS;aAC1G,CAAC;QACJ,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,eAAe,CAC7B,WAAmB;IAEnB,IAAI,MAAM,GAAyD,IAAI,CAAC;IACxE,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,IAAI,CAAC,OAAO;YAAE,SAAS;QACvB,IAAI,GAA4B,CAAC;QACjC,IAAI,CAAC;YAAC,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC;YAAC,SAAS;QAAC,CAAC;QACtD,IAAI,GAAG,CAAC,MAAM,CAAC,KAAK,gBAAgB,IAAI,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YACrD,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAA4B,CAAC;YACtD,MAAM,GAAG,GAAG,KAAK,CAAC,cAAc,CAAC,CAAC;YAClC,MAAM,IAAI,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC;YACpC,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;gBACxD,MAAM,GAAG,EAAE,WAAW,EAAE,GAAG,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;YACpD,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,cAAc,CAAC,KAAyB;IAC/C,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACpD,OAAO;QACL,KAAK,EAAE,KAAK,IAAI,SAAS;QACzB,OAAO,EAAE,GAAG,EAAE,GAAG,IAAI,0BAA0B;QAC/C,SAAS,EAAE,IAAI;KAChB,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CACjC,OAAe,EACf,UAAkB,EAClB,KAAyB;IAEzB,IAAI,CAAC;QACH,MAAM,SAAS,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;QAChD,IAAI,SAAS,KAAK,mBAAmB,EAAE,CAAC;YACtC,MAAM,GAAG,GAAG,YAAY,CAAC,IAAI,CAAC,UAAU,EAAE,cAAc,EAAE,cAAc,CAAC,EAAE,MAAM,CAAC,CAAC;YACnF,MAAM,MAAM,GAAG,oBAAoB,CAAC,GAAG,CAAC,CAAC;YACzC,IAAI,MAAM,EAAE,CAAC;gBACX,OAAO;oBACL,KAAK,EAAE,KAAK,IAAI,SAAS;oBACzB,OAAO,EAAE,MAAM,CAAC,OAAO;oBACvB,SAAS,EAAE,KAAK;oBAChB,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY;iBAClC,CAAC;YACJ,CAAC;YACD,OAAO,cAAc,CAAC,KAAK,CAAC,CAAC;QAC/B,CAAC;QACD,IAAI,SAAS,KAAK,aAAa,EAAE,CAAC;YAChC,MAAM,GAAG,GAAG,YAAY,CAAC,IAAI,CAAC,UAAU,EAAE,aAAa,EAAE,cAAc,CAAC,EAAE,MAAM,CAAC,CAAC;YAClF,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;YACpC,oEAAoE;YACpE,0DAA0D;YAC1D,IAAI,MAAM,IAAI,KAAK,EAAE,CAAC;gBACpB,MAAM,GAAG,GAAG,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;gBACvC,IAAI,GAAG,IAAI,IAAI,EAAE,CAAC;oBAChB,OAAO;wBACL,KAAK;wBACL,OAAO,EAAE,GAAG;wBACZ,SAAS,EAAE,KAAK;wBAChB,WAAW,EAAE,MAAM,CAAC,WAAW;wBAC/B,YAAY,EAAE,MAAM,CAAC,YAAY;qBAClC,CAAC;gBACJ,CAAC;YACH,CAAC;YACD,OAAO,cAAc,CAAC,KAAK,CAAC,CAAC;QAC/B,CAAC;QACD,OAAO,cAAc,CAAC,KAAK,CAAC,CAAC;IAC/B,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,cAAc,CAAC,KAAK,CAAC,CAAC;IAC/B,CAAC;AACH,CAAC"}

package/dist/store/store.d.ts

@@ -10,6 +10,17 @@ export interface ActivityEventInput {
     solverType?: string | null;
     outcome?: string | null;
     detail?: string | null;
+    credentialId?: string | null;
+    costUsdMicros?: number | null;
+    model?: string | null;
+    /** Projected AI units debited at claim time (issue #815). Estimates are the gate input; never recomputed. */
+    aiUnits?: number | null;
+    /** Lifecycle stamp on the per-request row: 'claimed' | 'claim_failed' | 'delivered'. */
+    claimStatus?: string | null;
+    /** USD estimate captured at claim time (micros). Distinct from `actualCostUsdMicros` filled on completion. */
+    estimatedCostUsdMicros?: number | null;
+    /** USD actually billed (micros) — filled by the completion path; null until then. */
+    actualCostUsdMicros?: number | null;
 }
 export interface ActivityEventRow {
     id: number;
@@ -21,6 +32,13 @@ export interface ActivityEventRow {
     solverType: string | null;
     outcome: string | null;
     detail: string | null;
+    credentialId: string | null;
+    costUsdMicros: number | null;
+    model: string | null;
+    aiUnits: number | null;
+    claimStatus: string | null;
+    estimatedCostUsdMicros: number | null;
+    actualCostUsdMicros: number | null;
 }
 export interface RewardClaimInput {
     ts: string;
@@ -148,6 +166,22 @@ export interface NetworkArtifactMetadataRow {
     lastUsedAt: string;
     peerCatalogId: string | null;
 }
+export interface Erc8004AnchorInput {
+    envelopeId: string;
+    envelopeCid: string;
+    contentKind: string;
+    metadataKey: string;
+    agentId: string;
+    chainId: number;
+    identityRegistryAddress: string;
+    txHash: string;
+    blockNumber: number | null;
+    payloadHex: string;
+    anchoredAt: number;
+}
+export interface Erc8004AnchorRow extends Erc8004AnchorInput {
+    id: number;
+}
 export type TaskPostingPolicyType = 'once_per_safe' | 'once_per_bucket' | 'interval';
 type LauncherTaskProjectionState = 'open' | 'claims-in-flight' | 'fully-claimed' | 'settled' | 'failed';
 export interface TaskPostRecord {
@@ -185,6 +219,8 @@ export declare class Store {
     private ensureTaskPostsTaskCoordinatorColumns;
     /** Older local DBs may have the projection table from before Task grouping fields landed. */
     private ensureEnvelopeProjectionColumns;
+    /** Older local DBs predate the per-credential spend-ledger columns on activity_events. */
+    private ensureActivityEventCostColumns;
     /**
      * Task-native startup ignores the retired request-first `restoration_intents`
      * table. Keep a one-time local marker when old in-flight rows are present so
@@ -194,6 +230,14 @@ export declare class Store {
     /** Idempotent: older DBs before idx_reward_claims_tx may lack the unique index. */
     private ensureRewardClaimsTxIndex;
     recordOwnActivity(requestId: string, role: 'created' | 'claimed' | 'delivered' | 'evaluated'): void;
+    /**
+     * Membership-only variant of `recordOwnActivity`: writes the
+     * `own_activity` row but does NOT emit a generic `activity_events`
+     * row. Used by paths that emit their own enriched activity event
+     * (e.g. issue #815's claim path attaches credentialId / aiUnits /
+     * estimatedCostUsdMicros / claimStatus to the row).
+     */
+    markOwnActivity(requestId: string, role: 'created' | 'claimed' | 'delivered' | 'evaluated'): void;
     isOwnActivity(requestId: string): boolean;
     setShutdownState(state: 'clean' | 'running'): void;
     getShutdownState(): string | null;
@@ -279,8 +323,106 @@ export declare class Store {
         since?: string;
         cursor?: string;
     }): ActivityEventRow[];
+    /**
+     * Total cost in micro-dollars recorded against a credential since the most
+     * recent UTC midnight. Backs the daily spend cap.
+     */
+    spentTodayMicros(credentialId: string, now?: Date): number;
+    /**
+     * Sum of `ai_units` for a credential within the current 6h UTC-aligned
+     * block (00:00 / 06:00 / 12:00 / 18:00 boundaries). Reads only rows whose
+     * `claim_status = 'claimed'` or `'delivered'` so failed-claim rows
+     * (`ai_units = 0`, `claim_status = 'claim_failed'`) don't muddy the sum
+     * even though their contribution is already zero.
+     *
+     * Issue #815. Backs the per-block AI-units ceiling gate.
+     */
+    aiUnitsThisBlock(credentialId: string, now?: Date): number;
+    /**
+     * Sum of `ai_units` for a credential within the trailing 7-day rolling
+     * window from `now`. Backs the per-week AI-units safety-net ceiling.
+     * Issue #815.
+     */
+    aiUnitsThisWeek(credentialId: string, now?: Date): number;
+    /**
+     * Actual-spend accumulator for the current 6h UTC block (issue #1004).
+     *
+     * Sums `COALESCE(actual_cost_usd_micros, estimated_cost_usd_micros, 0)`
+     * over rows whose `claim_status` is `'claimed'` or `'delivered'`:
+     *   - delivered rows contribute the real harvested cost (`actual_*`),
+     *   - in-flight claimed rows contribute their estimate so a burst of
+     *     concurrent claims cannot slip the cap before any of them deliver,
+     *   - failed claims (status `'claim_failed'`) are excluded.
+     *
+     * `estimated` is true iff the summed figure includes any estimate-backed
+     * cost: an in-flight `claimed` row with no `actual_cost_usd_micros` yet,
+     * OR a `delivered` row whose actual cost is itself a heuristic
+     * (`actual_cost_estimated = 1` — a telemetry-less harness such as Hermes).
+     * It is false only when every contributing row is harvested actual
+     * telemetry. The gate surfaces this so an estimate-backed figure is not
+     * presented as metered. Block boundaries mirror `aiUnitsThisBlock`.
+     */
+    usdMicrosThisBlock(credentialId: string, now?: Date): {
+        usdMicros: number;
+        estimated: boolean;
+    };
+    /**
+     * Actual-spend accumulator for the trailing 7-day rolling window from
+     * `now` (issue #1004). Same COALESCE + claim_status filter + `estimated`
+     * semantics as {@link usdMicrosThisBlock}.
+     */
+    usdMicrosThisWeek(credentialId: string, now?: Date): {
+        usdMicros: number;
+        estimated: boolean;
+    };
+    /** Shared COALESCE-sum + estimate-flag query for the USD accumulators. */
+    private sumUsdMicros;
+    /**
+     * True iff an `ai_units_cap_reached` row exists for the given
+     * (credentialId, window, blockId). Used by the daemon to hydrate the
+     * AI-units gate's in-memory pause memo across restarts so the
+     * "exactly one event per (credential, window, block-id)" guarantee
+     * holds across process boundaries (issue #815, finding 1).
+     *
+     * Lookup is by `credential_id` + `kind` + the `[block=...][window=...]`
+     * markers that `daemon.ts` embeds in the row's `detail` string.
+     */
+    hasAiUnitsCapReachedFor(credentialId: string, window: 'block' | 'week', blockId: string): boolean;
+    /**
+     * Mark the per-request `claimed` row as `delivered` and record
+     * `actual_cost_usd_micros` (issue #1004 — the gate's accumulator now
+     * reads this column via COALESCE, so a delivered row's real harvested
+     * cost replaces its claim-time estimate in the running total). The
+     * `ai_units` projection captured at claim time is intentionally NOT
+     * recomputed — it remains the per-task estimate for the legacy unit
+     * surfaces. For subscription credentials the resulting USD figure is a
+     * *proxy* budget, not an exact bound on the provider's plan quota.
+     *
+     * `actualCostEstimated` (issue #1004, AC4) records whether the actual
+     * cost is itself a heuristic — true for a telemetry-less harness such as
+     * Hermes whose `harvestHarnessUsage` falls back to an a-priori estimate,
+     * false when the figure is harvested telemetry. The gate reads it so a
+     * delivered-but-heuristic row reports `estimated: true` rather than being
+     * presented as metered. Idempotent: a no-op when no `claimed` row exists.
+     */
+    finalizeClaimDelivered(requestId: string, actualCostUsdMicros: number, actualCostEstimated: boolean): void;
     /** Newer events first, then ascending id for `jinn logs --follow` (oldest in batch printed first in caller). */
     getActivityEventsAfterId(afterId: number, limit: number): ActivityEventRow[];
+    /**
+     * Filtered, id-cursored page of activity events for the dedicated Events
+     * page. Newest-first.
+     *
+     * Cursors on `id` rather than `ts` so startup/shutdown rows with null
+     * timestamps remain reachable.
+     */
+    getActivityEventsPage(opts?: {
+        kinds?: string[];
+        outcome?: string;
+        requestId?: string;
+        beforeId?: number;
+        limit?: number;
+    }): ActivityEventRow[];
+    getActivityEventById(id: number): ActivityEventRow | null;
     getActivityCountsByKind(): Record<string, number>;
     getLastEventAtForService(serviceIndex: number): string | null;
     getActivityCountsForService(serviceIndex: number): Record<string, number>;
@@ -398,6 +540,51 @@ export declare class Store {
     }>;
     saveEnvelopeProjection(projection: EnvelopeProjection): void;
     queryEnvelopeProjections(query?: EnvelopeProjectionQuery): EnvelopeProjection[];
+    saveErc8004Anchor(input: Erc8004AnchorInput): void;
+    listErc8004AnchorsByEnvelopeCids(envelopeCids: readonly string[]): Erc8004AnchorRow[];
+    /**
+     * Upsert one held-out eval result (issue #818). PK is
+     * `(checkpoint_cid, slate_version, instance_id)` so a re-run overwrites.
+     * `passed` is null for `unscorable` rows.
+     */
+    recordEvalResult(args: EvalResultRecord): void;
+    /**
+     * Aggregate the eval results for a (checkpoint, slate version):
+     * `scorable` = rows with `unscorable = 0`; `passed` = scorable rows with
+     * `passed = 1`. Unscorable rows are counted separately and never enter the
+     * denominator. A checkpoint with no rows yields all-zero (the orchestrator
+     * reads this to detect a not-yet-evaluated parent).
+     */
+    getEvalAggregate(checkpoint_cid: string, slate_version: string): EvalAggregate;
+    /**
+     * Distinct `slate_hash` values recorded for a (checkpoint, slate version).
+     * The eval orchestrator reads this to detect slate-content drift under a
+     * stable version label — the held-out exam is only an honest before/after
+     * when the parent and child were scored on the SAME slate content (defeating
+     * confounder #1, task-selection). Empty when the checkpoint has no rows.
+     */
+    getEvalSlateHashes(checkpoint_cid: string, slate_version: string): string[];
+    /** Per-task eval results for a (checkpoint, slate version), ordered by instance_id. */
+    getEvalResults(checkpoint_cid: string, slate_version: string): EvalResultRow[];
     close(): void;
 }
+export interface EvalResultRecord {
+    checkpoint_cid: string;
+    slate_hash: string;
+    slate_version: string;
+    instance_id: string;
+    /** Pass/fail; ignored (stored NULL) when `unscorable` is true. */
+    passed: boolean | null;
+    unscorable: boolean;
+    code_digest: string;
+    run_at_ms: number;
+    test_log_excerpt?: string | null;
+}
+/** A persisted eval result read back from the store (same shape as the record written). */
+export type EvalResultRow = EvalResultRecord;
+export interface EvalAggregate {
+    passed: number;
+    scorable: number;
+    unscorable: number;
+}
 export {};