@tangle-network/agent-runtime 0.105.0 → 0.106.1

package/dist/agent.d.ts

@@ -1,13 +1,11 @@
-import * as _tangle_network_agent_eval from '@tangle-network/agent-eval';
-import { TraceAnalystKindSpec } from '@tangle-network/agent-eval';
-import { R as RuntimeStreamEvent } from './types-BwoZWq-i.js';
-import { A as AgentSurfaces } from './improvement-adapter-HAZz-7vK.js';
-export { C as CreateSurfaceImprovementProposerOptions, D as DraftPatchInput, a as DraftPatchOutput, R as ResolvedSurface, S as SurfaceImprovementEdit, b as SurfaceValidationIssue, c as createSurfaceImprovementProposer, r as renderSurfaceIssues, d as resolveSubjectPath, v as validateSurfaces } from './improvement-adapter-HAZz-7vK.js';
-import { AgentProfile, AgentProfileFileMount, AgentProfileMcpServer } from '@tangle-network/agent-interface';
-import { SandboxEvent } from '@tangle-network/sandbox';
-import { a as SandboxClient, O as OutputAdapter, A as AgentRunSpec } from './types-BCoemcXU.js';
-import './types-CmYCMbFT.js';
-
+import { x as RuntimeStreamEvent } from "./types-C9j4qg6l.js";
+import { a as createSurfaceImprovementProposer, c as SurfaceValidationIssue, d as validateSurfaces, i as SurfaceImprovementEdit, l as renderSurfaceIssues, n as DraftPatchInput, o as AgentSurfaces, r as DraftPatchOutput, s as ResolvedSurface, t as CreateSurfaceImprovementProposerOptions, u as resolveSubjectPath } from "./improvement-adapter-Cr5TBSji.js";
+import { E as SandboxClient, t as AgentRunSpec, w as OutputAdapter } from "./types-BevOjfTY.js";
+import "./index-DruSKs3Z.js";
+import { TraceAnalystKindSpec } from "@tangle-network/agent-eval";
+import { AgentProfile as AgentProfile$1, AgentProfileFileMount, AgentProfileMcpServer } from "@tangle-network/agent-interface";
+import { SandboxEvent } from "@tangle-network/sandbox";
+//#region src/agent/define-agent.d.ts
 /**
  * The full agent manifest. Each agent ships ONE of these.
  *
@@ -20,145 +18,145 @@ import './types-CmYCMbFT.js';
  *     scorers and emitted into the trace.
  */
 interface AgentManifest<TPersona = unknown, TRunOutput = unknown> {
-    /**
-     * Stable identifier — used as `projectId` in traces, as the analyst
-     * loop's `runId` prefix, and as the namespace under which findings
-     * are persisted. MUST match the agent's repo name to keep
-     * cross-repo telemetry joinable.
-     */
-    id: string;
-    /**
-     * Filesystem root the substrate resolves surface paths against.
-     * Typically `process.cwd()` or a fixed absolute path. Use an
-     * absolute path when the agent's tests may run from subdirectories
-     * (vitest sometimes shifts cwd).
-     */
-    repoRoot: string;
-    /**
-     * Map of mutable surfaces the self-improvement loop can edit. See
-     * `AgentSurfaces` — required: `systemPrompt`, `tools`, `rubric`,
-     * `knowledge`, `personas`. Optional: `scaffolding`, `memory`, `rag`,
-     * `outputSchema`.
-     *
-     * Every required path is validated at `defineAgent` time. Missing
-     * paths throw with the full list of offenders.
-     */
-    surfaces: AgentSurfaces;
-    /**
-     * Rubric the substrate uses to score each run. Dimensions × weights
-     * × judges. The substrate computes the weighted composite and
-     * stamps it into the RunRecord.
-     */
-    rubric: AgentRubric<TRunOutput>;
-    /**
-     * Runtime adapter — how the substrate INVOKES the agent against a
-     * persona. The `act` function takes a persona + a context (with the
-     * tracer the substrate threads through for span emission) and
-     * returns the run output the rubric will score.
-     *
-     * The agent's existing production runtime goes in here; the
-     * substrate is intentionally thin around it.
-     */
-    runtime: AgentRuntime<TPersona, TRunOutput>;
-    /**
-     * Persona discovery — the substrate loads personas via this function
-     * at eval start. Can read from `surfaces.personas`, an API, or be
-     * hardcoded. The substrate calls it once per `runAgentEval` call;
-     * persona ordering is preserved.
-     */
-    personas: () => Promise<ReadonlyArray<TPersona>>;
-    /**
-     * Analyst kinds the substrate runs against each persona's trace.
-     * Defaults to `DEFAULT_TRACE_ANALYST_KINDS` from agent-eval. Per-agent
-     * authors can prune (e.g. skip `knowledge-poisoning` when there's no
-     * knowledge base) or extend (custom domain kinds).
-     *
-     * Empty array disables the loop — useful for `pnpm eval --no-analyst`.
-     */
-    analystKinds: ReadonlyArray<TraceAnalystKindSpec>;
-    /**
-     * Analyst LLM configuration. The substrate uses these for all four
-     * kinds (override per-kind via `analystKinds` if needed).
-     */
-    analyst: AnalystConfig;
+  /**
+   * Stable identifier — used as `projectId` in traces, as the analyst
+   * loop's `runId` prefix, and as the namespace under which findings
+   * are persisted. MUST match the agent's repo name to keep
+   * cross-repo telemetry joinable.
+   */
+  id: string;
+  /**
+   * Filesystem root the substrate resolves surface paths against.
+   * Typically `process.cwd()` or a fixed absolute path. Use an
+   * absolute path when the agent's tests may run from subdirectories
+   * (vitest sometimes shifts cwd).
+   */
+  repoRoot: string;
+  /**
+   * Map of mutable surfaces the self-improvement loop can edit. See
+   * `AgentSurfaces` — required: `systemPrompt`, `tools`, `rubric`,
+   * `knowledge`, `personas`. Optional: `scaffolding`, `memory`, `rag`,
+   * `outputSchema`.
+   *
+   * Every required path is validated at `defineAgent` time. Missing
+   * paths throw with the full list of offenders.
+   */
+  surfaces: AgentSurfaces;
+  /**
+   * Rubric the substrate uses to score each run. Dimensions × weights
+   * × judges. The substrate computes the weighted composite and
+   * stamps it into the RunRecord.
+   */
+  rubric: AgentRubric<TRunOutput>;
+  /**
+   * Runtime adapter — how the substrate INVOKES the agent against a
+   * persona. The `act` function takes a persona + a context (with the
+   * tracer the substrate threads through for span emission) and
+   * returns the run output the rubric will score.
+   *
+   * The agent's existing production runtime goes in here; the
+   * substrate is intentionally thin around it.
+   */
+  runtime: AgentRuntime<TPersona, TRunOutput>;
+  /**
+   * Persona discovery — the substrate loads personas via this function
+   * at eval start. Can read from `surfaces.personas`, an API, or be
+   * hardcoded. The substrate calls it once per `runAgentEval` call;
+   * persona ordering is preserved.
+   */
+  personas: () => Promise<ReadonlyArray<TPersona>>;
+  /**
+   * Analyst kinds the substrate runs against each persona's trace.
+   * Defaults to `DEFAULT_TRACE_ANALYST_KINDS` from agent-eval. Per-agent
+   * authors can prune (e.g. skip `knowledge-poisoning` when there's no
+   * knowledge base) or extend (custom domain kinds).
+   *
+   * Empty array disables the loop — useful for `pnpm eval --no-analyst`.
+   */
+  analystKinds: ReadonlyArray<TraceAnalystKindSpec>;
+  /**
+   * Analyst LLM configuration. The substrate uses these for all four
+   * kinds (override per-kind via `analystKinds` if needed).
+   */
+  analyst: AnalystConfig;
 }
 interface AgentRubric<TRunOutput> {
-    /** Dimensions composing the weighted score. Weights sum to 1.0 by convention. */
-    dimensions: ReadonlyArray<RubricDimension<TRunOutput>>;
-    /**
-     * Optional judges layered on top of deterministic dimensions. Each
-     * judge returns a score per dimension; the substrate averages judges
-     * (mean by default) for the LLM contribution.
-     */
-    judges?: ReadonlyArray<JudgeConfig<TRunOutput>>;
+  /** Dimensions composing the weighted score. Weights sum to 1.0 by convention. */
+  dimensions: ReadonlyArray<RubricDimension<TRunOutput>>;
+  /**
+   * Optional judges layered on top of deterministic dimensions. Each
+   * judge returns a score per dimension; the substrate averages judges
+   * (mean by default) for the LLM contribution.
+   */
+  judges?: ReadonlyArray<JudgeConfig<TRunOutput>>;
 }
 interface RubricDimension<TRunOutput> {
-    /** Unique identifier — appears in finding subjects (`rubric:<id>`). */
-    id: string;
-    /** 0..1 — weight in the composite. */
-    weight: number;
-    /**
-     * Deterministic scorer: given the persona + run output, returns a
-     * 0..1 score. The substrate sums weight × score across dimensions
-     * for the deterministic composite; judges supplement subjective dims.
-     */
-    score: (input: {
-        persona: unknown;
-        output: TRunOutput;
-    }) => number;
-    /** Optional human-readable label for reports. */
-    label?: string;
+  /** Unique identifier — appears in finding subjects (`rubric:<id>`). */
+  id: string;
+  /** 0..1 — weight in the composite. */
+  weight: number;
+  /**
+   * Deterministic scorer: given the persona + run output, returns a
+   * 0..1 score. The substrate sums weight × score across dimensions
+   * for the deterministic composite; judges supplement subjective dims.
+   */
+  score: (input: {
+    persona: unknown;
+    output: TRunOutput;
+  }) => number;
+  /** Optional human-readable label for reports. */
+  label?: string;
 }
 interface JudgeConfig<TRunOutput> {
-    /** Judge identifier — appears in trace spans + manifest. */
-    id: string;
-    /** Model snapshot to invoke. Pin the snapshot (`claude-sonnet-4-6@2025-04-15`); the validator rejects bare aliases. */
-    model: string;
-    /** Dimensions this judge scores. */
-    dimensions: ReadonlyArray<string>;
-    /**
-     * Optional rubric anchors — text examples the judge sees as a
-     * few-shot prompt to calibrate. STRONGLY recommended for subjective
-     * dimensions; required by the calibration gate (Pearson ≥0.7).
-     */
-    anchors?: ReadonlyArray<{
-        input: string;
-        output: TRunOutput;
-        expected: Record<string, number>;
-    }>;
+  /** Judge identifier — appears in trace spans + manifest. */
+  id: string;
+  /** Model snapshot to invoke. Pin the snapshot (`claude-sonnet-4-6@2025-04-15`); the validator rejects bare aliases. */
+  model: string;
+  /** Dimensions this judge scores. */
+  dimensions: ReadonlyArray<string>;
+  /**
+   * Optional rubric anchors — text examples the judge sees as a
+   * few-shot prompt to calibrate. STRONGLY recommended for subjective
+   * dimensions; required by the calibration gate (Pearson ≥0.7).
+   */
+  anchors?: ReadonlyArray<{
+    input: string;
+    output: TRunOutput;
+    expected: Record<string, number>;
+  }>;
 }
 interface AgentRuntime<TPersona, TRunOutput> {
-    /**
-     * Invoke the agent against one persona. Returns BOTH:
-     *   - `events`: an `AsyncIterable<RuntimeStreamEvent>` the chat-centric
-     *     product consumes verbatim (SSE / WebSocket / inline render).
-     *     **Streaming is mandatory — never collapse this to a single Promise.**
-     *     The agent's existing `runChatTurn` (or equivalent async generator)
-     *     plugs in here directly.
-     *   - `output`: a `Promise<TRunOutput>` resolved AFTER the event stream
-     *     drains. The eval substrate awaits this for rubric scoring; chat
-     *     products usually ignore it (they already rendered incrementally).
-     *
-     * Implementation contract:
-     *   1. `act` MUST return immediately (synchronous construction of the
-     *      `events` iterator + the `output` promise).
-     *   2. Iterating `events` drives the underlying LLM/tool calls — the
-     *      caller chooses when to consume.
-     *   3. `output` resolves only after the iterator yields its terminal
-     *      event (typically `task_end`); see `collectAgentRun` helper.
-     *
-     * `ctx.emitter` is the substrate-threaded `TraceEmitter` — runtimes
-     * SHOULD record LLM/tool spans through it for capture integrity.
-     * `ctx.deadlineMs` is wall-clock; the runtime SHOULD honour for graceful
-     * cancel. `ctx.signal` is the standard abort signal.
-     */
-    act: (persona: TPersona, ctx: AgentRunContext) => AgentRunInvocation<TRunOutput>;
+  /**
+   * Invoke the agent against one persona. Returns BOTH:
+   *   - `events`: an `AsyncIterable<RuntimeStreamEvent>` the chat-centric
+   *     product consumes verbatim (SSE / WebSocket / inline render).
+   *     **Streaming is mandatory — never collapse this to a single Promise.**
+   *     The agent's existing `runChatTurn` (or equivalent async generator)
+   *     plugs in here directly.
+   *   - `output`: a `Promise<TRunOutput>` resolved AFTER the event stream
+   *     drains. The eval substrate awaits this for rubric scoring; chat
+   *     products usually ignore it (they already rendered incrementally).
+   *
+   * Implementation contract:
+   *   1. `act` MUST return immediately (synchronous construction of the
+   *      `events` iterator + the `output` promise).
+   *   2. Iterating `events` drives the underlying LLM/tool calls — the
+   *      caller chooses when to consume.
+   *   3. `output` resolves only after the iterator yields its terminal
+   *      event (typically `task_end`); see `collectAgentRun` helper.
+   *
+   * `ctx.emitter` is the substrate-threaded `TraceEmitter` — runtimes
+   * SHOULD record LLM/tool spans through it for capture integrity.
+   * `ctx.deadlineMs` is wall-clock; the runtime SHOULD honour for graceful
+   * cancel. `ctx.signal` is the standard abort signal.
+   */
+  act: (persona: TPersona, ctx: AgentRunContext) => AgentRunInvocation<TRunOutput>;
 }
 interface AgentRunInvocation<TRunOutput> {
-    /** Live stream of typed runtime events. Consumed by chat UX directly. */
-    events: AsyncIterable<RuntimeStreamEvent>;
-    /** Final structured output the rubric scores. Resolves after `events` drains. */
-    output: Promise<TRunOutput>;
+  /** Live stream of typed runtime events. Consumed by chat UX directly. */
+  events: AsyncIterable<RuntimeStreamEvent>;
+  /** Final structured output the rubric scores. Resolves after `events` drains. */
+  output: Promise<TRunOutput>;
 }
 /**
  * Stub for agents whose `runtime.act` is not yet wired to the substrate's
@@ -181,38 +179,38 @@ declare function unimplementedAgentRun<TRunOutput = unknown>(reason?: string): A
  * directly in the chat surface.
  */
 declare function collectAgentRun<TRunOutput>(invocation: AgentRunInvocation<TRunOutput>): Promise<{
-    events: ReadonlyArray<RuntimeStreamEvent>;
-    output: TRunOutput;
+  events: ReadonlyArray<RuntimeStreamEvent>;
+  output: TRunOutput;
 }>;
 interface AgentRunContext {
-    /** Substrate-managed trace emitter. */
-    emitter: _tangle_network_agent_eval.TraceEmitter;
-    /** Stable run id for this persona × variant cell. */
-    runId: string;
-    /** Variant the runtime is exercising (e.g. `'baseline'`, `'source-grounded'`). */
-    variantId?: string;
-    /** Wall-clock deadline (epoch ms). The runtime SHOULD honour for graceful cancel. */
-    deadlineMs?: number;
-    /** Optional abort signal. */
-    signal?: AbortSignal;
+  /** Substrate-managed trace emitter. */
+  emitter: import('@tangle-network/agent-eval').TraceEmitter;
+  /** Stable run id for this persona × variant cell. */
+  runId: string;
+  /** Variant the runtime is exercising (e.g. `'baseline'`, `'source-grounded'`). */
+  variantId?: string;
+  /** Wall-clock deadline (epoch ms). The runtime SHOULD honour for graceful cancel. */
+  deadlineMs?: number;
+  /** Optional abort signal. */
+  signal?: AbortSignal;
 }
 interface AnalystConfig {
-    /** Model the analyst kinds use. Override per-kind via `analystKinds[i].cost.models`. */
-    model: string;
-    /** Optional total budget across all kinds for one run. Substrate enforces via `BudgetGuard`. */
-    budgetUsd?: number;
-    /** Backend hint for the AxAIService factory — same shape every kind uses. */
-    backend?: {
-        name?: 'openai' | 'router';
-        apiKey?: string;
-        baseUrl?: string;
-    };
+  /** Model the analyst kinds use. Override per-kind via `analystKinds[i].cost.models`. */
+  model: string;
+  /** Optional total budget across all kinds for one run. Substrate enforces via `BudgetGuard`. */
+  budgetUsd?: number;
+  /** Backend hint for the AxAIService factory — same shape every kind uses. */
+  backend?: {
+    name?: 'openai' | 'router';
+    apiKey?: string;
+    baseUrl?: string;
+  };
 }
 /** Thrown when `defineAgent` finds a required surface missing on disk. */
 declare class AgentManifestError extends Error {
-    readonly agentId: string;
-    readonly issues: ReadonlyArray<unknown>;
-    constructor(message: string, agentId: string, issues?: ReadonlyArray<unknown>);
+  readonly agentId: string;
+  readonly issues: ReadonlyArray<unknown>;
+  constructor(message: string, agentId: string, issues?: ReadonlyArray<unknown>);
 }
 /**
  * Construct a validated agent manifest. Throws `AgentManifestError`
@@ -228,7 +226,8 @@ declare class AgentManifestError extends Error {
  * the call site.
  */
 declare function defineAgent<TPersona = unknown, TRunOutput = unknown>(manifest: AgentManifest<TPersona, TRunOutput>): AgentManifest<TPersona, TRunOutput>;
-
+//#endregion
+//#region src/agent/profile-materialization.d.ts
 /** Known AgentProfile axes a run path may or may not carry into execution. */
 declare const AGENT_PROFILE_MATERIALIZATION_AXES: readonly ["identity", "name", "model", "prompt", "systemPrompt", "instructions", "resources", "files", "resourceInstructions", "skills", "resourceTools", "resourceAgents", "commands", "tools", "permissions", "mcp", "mcpConnections", "connections", "subagents", "hooks", "modes", "confidential", "metadata", "extensions"];
 type KnownAgentProfileMaterializationAxis = (typeof AGENT_PROFILE_MATERIALIZATION_AXES)[number];
@@ -236,32 +235,32 @@ type KnownAgentProfileMaterializationAxis = (typeof AGENT_PROFILE_MATERIALIZATIO
 type AgentProfileMaterializationAxis = KnownAgentProfileMaterializationAxis | `custom:${string}`;
 /** Declares which AgentProfile axes a concrete run path really carries. */
 interface ProfileMaterializationContract {
-    /** Human-readable run path, e.g. `createSandboxAct` or `prompt-only-message`. */
-    name: string;
-    /** Profile axes this run path actually carries into execution. */
-    axes: readonly AgentProfileMaterializationAxis[];
+  /** Human-readable run path, e.g. `createSandboxAct` or `prompt-only-message`. */
+  name: string;
+  /** Profile axes this run path actually carries into execution. */
+  axes: readonly AgentProfileMaterializationAxis[];
 }
 /** One changed AgentProfile axis that would be dropped by a run path. */
 interface ProfileMaterializationIssue {
-    contract: string;
-    axis: AgentProfileMaterializationAxis;
-    reason: 'unsupported-axis';
-    supportedAxes: readonly AgentProfileMaterializationAxis[];
+  contract: string;
+  axis: AgentProfileMaterializationAxis;
+  reason: 'unsupported-axis';
+  supportedAxes: readonly AgentProfileMaterializationAxis[];
 }
 /** Input for declaring a run path's profile-axis support. */
 interface DefineProfileMaterializationContractOptions {
-    name: string;
-    axes: readonly AgentProfileMaterializationAxis[];
+  name: string;
+  axes: readonly AgentProfileMaterializationAxis[];
 }
 /** Input for checking a candidate diff against a run path. */
 interface ValidateProfileMaterializationOptions {
-    contract: ProfileMaterializationContract;
-    changedAxes: readonly AgentProfileMaterializationAxis[];
+  contract: ProfileMaterializationContract;
+  changedAxes: readonly AgentProfileMaterializationAxis[];
 }
 /** Input for throwing on dropped profile axes. */
 interface AssertProfileMaterializationOptions extends ValidateProfileMaterializationOptions {
-    /** Extra label included in the thrown error, usually the caller or run id. */
-    context?: string;
+  /** Extra label included in the thrown error, usually the caller or run id. */
+  context?: string;
 }
 /** Materialization contract for `createSandboxAct`, which forwards the full AgentProfile. */
 declare const sandboxActProfileMaterialization: ProfileMaterializationContract;
@@ -277,64 +276,46 @@ declare function validateProfileMaterialization(options: ValidateProfileMaterial
 declare function assertProfileMaterialization(options: AssertProfileMaterializationOptions): void;
 /** Format profile-axis drop issues into a concise operator-facing error. */
 declare function renderProfileMaterializationIssues(issues: readonly ProfileMaterializationIssue[], context?: string): string;
-
-/**
- * Sandbox bridge for `AgentRuntime.act` — prod-faithful eval execution.
- *
- * The point of this adapter is parity: the eval substrate must run the agent
- * through the SAME profile the production chat turn uses, or scorecard numbers
- * grade a profile that never ships. `createSandboxAct` boots a sandbox with the
- * agent's profile (the caller's `baseProfile`, with optional per-persona
- * overrides) through the loop kernel's own {@link createSandboxForSpec},
- * streams the `streamPrompt` events mapped to the `RuntimeStreamEvent`
- * vocabulary, and resolves the `OutputAdapter`-parsed output for rubric scoring
- * — satisfying the `act` streaming contract with one code path shared by chat
- * and eval.
- *
- * Agents with a bespoke streaming chat turn should wire THAT into `act`
- * directly (the contract is designed for it); this adapter is the default for
- * agents whose turn is a plain prod-profile sandbox dispatch — notably the
- * agents agent-builder generates.
- */
-
+//#endregion
+//#region src/agent/sandbox-act.d.ts
 /** Per-persona profile-merge slots applied over the base profile (§1.5: the caller authors the
  *  per-persona profile). Each slot overlays the base; an absent slot leaves the base untouched. */
 interface SandboxActComposeOverrides {
-    /** Replace the base profile's system prompt (e.g. a workspace-augmented prompt). */
-    systemPrompt?: string;
-    /** Extra file mounts layered after the base profile's `resources.files`. */
-    extraFiles?: AgentProfileFileMount[];
-    /** Override the profile `name`. Defaults to the base profile's name. */
-    name?: string;
-    /** Box built-in tool ON/OFF flags merged over the base profile's `tools` (overlay wins per key). */
-    tools?: Record<string, boolean>;
-    /** MCP connections merged over the base profile's `mcp` (overlay wins per key). */
-    mcpConnections?: Record<string, AgentProfileMcpServer>;
+  /** Replace the base profile's system prompt (e.g. a workspace-augmented prompt). */
+  systemPrompt?: string;
+  /** Extra file mounts layered after the base profile's `resources.files`. */
+  extraFiles?: AgentProfileFileMount[];
+  /** Override the profile `name`. Defaults to the base profile's name. */
+  name?: string;
+  /** Box built-in tool ON/OFF flags merged over the base profile's `tools` (overlay wins per key). */
+  tools?: Record<string, boolean>;
+  /** MCP connections merged over the base profile's `mcp` (overlay wins per key). */
+  mcpConnections?: Record<string, AgentProfileMcpServer>;
 }
 interface CreateSandboxActOptions<TPersona, TRunOutput> {
-    /** Canonical agent profile — the same one the prod chat turn uses. */
-    baseProfile: AgentProfile;
-    /** Sandbox client used to boot the per-run sandbox. */
-    sandboxClient: SandboxClient;
-    /** Persona → prompt. Pure; the eval cell's input. */
-    buildPrompt: (persona: TPersona) => string;
-    /** Sandbox event stream → typed output the rubric scores. */
-    output: OutputAdapter<TRunOutput>;
-    /**
-     * Per-persona profile overrides (workspace-augmented system prompt, extra
-     * file mounts, tool flags, MCP connections). Overlaid onto `baseProfile`.
-     */
-    compose?: (persona: TPersona) => SandboxActComposeOverrides;
-    /** Sandbox-SDK overrides forwarded to `createSandboxForSpec`. */
-    sandboxOverrides?: AgentRunSpec<unknown>['sandboxOverrides'];
-    /** Optional changed axes the caller expects this path to carry. */
-    requiredProfileAxes?: readonly AgentProfileMaterializationAxis[];
-    /** Stable run name surfaced in mapped `llm_call` events. */
-    name?: string;
-    /** Override the `SandboxEvent → RuntimeStreamEvent` mapper. */
-    mapEvent?: (event: SandboxEvent, opts: {
-        agentRunName?: string;
-    }) => RuntimeStreamEvent | undefined;
+  /** Canonical agent profile — the same one the prod chat turn uses. */
+  baseProfile: AgentProfile$1;
+  /** Sandbox client used to boot the per-run sandbox. */
+  sandboxClient: SandboxClient;
+  /** Persona → prompt. Pure; the eval cell's input. */
+  buildPrompt: (persona: TPersona) => string;
+  /** Sandbox event stream → typed output the rubric scores. */
+  output: OutputAdapter<TRunOutput>;
+  /**
+   * Per-persona profile overrides (workspace-augmented system prompt, extra
+   * file mounts, tool flags, MCP connections). Overlaid onto `baseProfile`.
+   */
+  compose?: (persona: TPersona) => SandboxActComposeOverrides;
+  /** Sandbox-SDK overrides forwarded to `createSandboxForSpec`. */
+  sandboxOverrides?: AgentRunSpec<unknown>['sandboxOverrides'];
+  /** Optional changed axes the caller expects this path to carry. */
+  requiredProfileAxes?: readonly AgentProfileMaterializationAxis[];
+  /** Stable run name surfaced in mapped `llm_call` events. */
+  name?: string;
+  /** Override the `SandboxEvent → RuntimeStreamEvent` mapper. */
+  mapEvent?: (event: SandboxEvent, opts: {
+    agentRunName?: string;
+  }) => RuntimeStreamEvent | undefined;
 }
 /**
  * Build an `AgentRuntime.act` implementation backed by a single prod-profile
@@ -343,5 +324,6 @@ interface CreateSandboxActOptions<TPersona, TRunOutput> {
  * resolves only after the iterator drains.
  */
 declare function createSandboxAct<TPersona, TRunOutput>(options: CreateSandboxActOptions<TPersona, TRunOutput>): (persona: TPersona, ctx: AgentRunContext) => AgentRunInvocation<TRunOutput>;
-
-export { AGENT_PROFILE_MATERIALIZATION_AXES, type AgentManifest, AgentManifestError, type AgentProfileMaterializationAxis, type AgentRubric, type AgentRunContext, type AgentRunInvocation, type AgentRuntime, AgentSurfaces, type AnalystConfig, type AssertProfileMaterializationOptions, type CreateSandboxActOptions, type DefineProfileMaterializationContractOptions, type JudgeConfig, type KnownAgentProfileMaterializationAxis, type ProfileMaterializationContract, type ProfileMaterializationIssue, type RubricDimension, type ValidateProfileMaterializationOptions, assertProfileMaterialization, collectAgentRun, createSandboxAct, defineAgent, defineProfileMaterializationContract, promptOnlyProfileMaterialization, promptResourceProfileMaterialization, renderProfileMaterializationIssues, sandboxActProfileMaterialization, unimplementedAgentRun, validateProfileMaterialization };
+//#endregion
+export { AGENT_PROFILE_MATERIALIZATION_AXES, type AgentManifest, AgentManifestError, type AgentProfileMaterializationAxis, type AgentRubric, type AgentRunContext, type AgentRunInvocation, type AgentRuntime, type AgentSurfaces, type AnalystConfig, type AssertProfileMaterializationOptions, type CreateSandboxActOptions, type CreateSurfaceImprovementProposerOptions, type DefineProfileMaterializationContractOptions, type DraftPatchInput, type DraftPatchOutput, type JudgeConfig, type KnownAgentProfileMaterializationAxis, type ProfileMaterializationContract, type ProfileMaterializationIssue, type ResolvedSurface, type RubricDimension, type SandboxActComposeOverrides, type SurfaceImprovementEdit, type SurfaceValidationIssue, type ValidateProfileMaterializationOptions, assertProfileMaterialization, collectAgentRun, createSandboxAct, createSurfaceImprovementProposer, defineAgent, defineProfileMaterializationContract, promptOnlyProfileMaterialization, promptResourceProfileMaterialization, renderProfileMaterializationIssues, renderSurfaceIssues, resolveSubjectPath, sandboxActProfileMaterialization, unimplementedAgentRun, validateProfileMaterialization, validateSurfaces };
+//# sourceMappingURL=agent.d.ts.map