@tangle-network/agent-eval 0.94.0 → 0.95.0

package/dist/muffled-gate-scanner.d.ts

@@ -0,0 +1,102 @@
+/**
+ * muffled-gate-scanner — test helper that greps consumer source for
+ * gate + measurement anti-patterns and fails with file:line locations.
+ *
+ * Named pattern lives at starter-foundry's `.evolve/patterns/muffled-gate.md`;
+ * same shape applies to every consumer (a gate that should fail loud
+ * returns silent success; a metric that should emit a real number
+ * reports noise/empty).
+ *
+ * Usage (in a consumer project's test file):
+ *
+ *   import { scanForMuffledGates, DEFAULT_FINDERS } from '@tangle-network/agent-eval'
+ *
+ *   test('no muffled gates in eval surface', () => {
+ *     const findings = scanForMuffledGates({
+ *       repoRoot: process.cwd(),
+ *       scanFiles: ['src/eval/scaffold.ts', 'scripts/promote.mjs'],
+ *       finders: DEFAULT_FINDERS,
+ *     })
+ *     if (findings.length) assert.fail(formatFindings(findings))
+ *   })
+ *
+ * Customize by passing your own `finders` — each finder is
+ * `(file, text) => Finding[]` and runs per-file.
+ *
+ * Escape hatch: any line containing `muffle-ok:` is excluded from all
+ * finders, letting consumers opt a legitimate fallback out explicitly.
+ */
+export interface MuffledFinding {
+    file: string;
+    line: number;
+    lineText: string;
+    pattern: string;
+}
+export type MuffledFinder = (file: string, text: string) => MuffledFinding[];
+export interface ScanOptions {
+    /** Absolute path to the repo root. */
+    repoRoot: string;
+    /** Explicit file list (paths relative to repoRoot) for context-specific finders. */
+    scanFiles: string[];
+    /**
+     * Auto-derived scan: walk these dirs for files matching importGlob + the
+     * string `importsContain` and run the universal finders on them. Pattern
+     * from starter-foundry H4 (research/decisions/001) — catches new files
+     * with agent-eval import that would otherwise escape context-specific
+     * scan lists.
+     */
+    autoDerive?: {
+        roots: string[];
+        extensions: RegExp;
+        importsContain: string;
+        universalFinders: MuffledFinder[];
+    };
+    /** Per-file finders (context-specific patterns). */
+    finders: MuffledFinder[];
+}
+/**
+ * Default finder: `command || true` in a testCommand/setupCommand/cmd/command
+ * string. Swallows exit codes.
+ */
+export declare const findFallbackToPass: MuffledFinder;
+/**
+ * `testCommand: 'true'` literal silent-pass — an unknown-language dispatch
+ * arm that returns a no-op instead of throwing.
+ */
+export declare const findLiteralTruePass: MuffledFinder;
+/**
+ * `new SubprocessSandboxDriver({ cwd: ... })` — constructor arg silently
+ * dropped in agent-eval <0.7.1. 0.7.1+ honors as fallback, but the form
+ * still invites confusion; prefer `new SubprocessSandboxDriver()` with
+ * cwd in the per-call HarnessConfig.
+ */
+export declare const findConstructorCwdDropped: MuffledFinder;
+/**
+ * `if (!expected) return true` — matcher auto-passes when ground truth is
+ * absent. Inflates accuracy metrics for scenarios without expectations.
+ */
+export declare const findAutoMatchNoExpectation: MuffledFinder;
+/**
+ * `if (p.skipped) return true` — skip-counts-as-pass in quality scorers.
+ * Use three-valued `true | false | 'skipped'` return + explicit partial
+ * credit instead.
+ */
+export declare const findSkipCountsAsPass: MuffledFinder;
+/**
+ * The canonical default bundle. Callers can import these individually,
+ * replace them, or append custom finders for project-specific patterns.
+ */
+export declare const DEFAULT_FINDERS: MuffledFinder[];
+/** Finders that should run on EVERY file with the target import, not just SCAN_FILES. */
+export declare const UNIVERSAL_FINDERS: MuffledFinder[];
+/**
+ * Run all finders against the configured files. Returns a flat list of
+ * findings. Callers format + assert as they prefer.
+ */
+export declare function scanForMuffledGates(opts: ScanOptions): MuffledFinding[];
+/**
+ * Format findings into a single assert.fail-ready message. Each finding
+ * carries file:line + pattern name + the offending line.
+ */
+export declare function formatFindings(findings: MuffledFinding[]): string;
+//# sourceMappingURL=muffled-gate-scanner.d.ts.map

package/dist/muffled-gate-scanner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"muffled-gate-scanner.d.ts","sourceRoot":"","sources":["../src/muffled-gate-scanner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAKH,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,EAAE,MAAM,CAAA;IAChB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,MAAM,aAAa,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,KAAK,cAAc,EAAE,CAAA;AAE5E,MAAM,WAAW,WAAW;IAC1B,sCAAsC;IACtC,QAAQ,EAAE,MAAM,CAAA;IAChB,oFAAoF;IACpF,SAAS,EAAE,MAAM,EAAE,CAAA;IACnB;;;;;;OAMG;IACH,UAAU,CAAC,EAAE;QACX,KAAK,EAAE,MAAM,EAAE,CAAA;QACf,UAAU,EAAE,MAAM,CAAA;QAClB,cAAc,EAAE,MAAM,CAAA;QACtB,gBAAgB,EAAE,aAAa,EAAE,CAAA;KAClC,CAAA;IACD,oDAAoD;IACpD,OAAO,EAAE,aAAa,EAAE,CAAA;CACzB;AAeD;;;GAGG;AACH,eAAO,MAAM,kBAAkB,EAAE,aAkBhC,CAAA;AAED;;;GAGG;AACH,eAAO,MAAM,mBAAmB,EAAE,aAkBjC,CAAA;AAED;;;;;GAKG;AACH,eAAO,MAAM,yBAAyB,EAAE,aAmBvC,CAAA;AAED;;;GAGG;AACH,eAAO,MAAM,0BAA0B,EAAE,aAkBxC,CAAA;AAED;;;;GAIG;AACH,eAAO,MAAM,oBAAoB,EAAE,aAkBlC,CAAA;AAED;;;GAGG;AACH,eAAO,MAAM,eAAe,EAAE,aAAa,EAK1C,CAAA;AAED,yFAAyF;AACzF,eAAO,MAAM,iBAAiB,EAAE,aAAa,EAAgC,CAAA;AAmD7E;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,WAAW,GAAG,cAAc,EAAE,CA+BvE;AAED;;;GAGG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,cAAc,EAAE,GAAG,MAAM,CAQjE"}

package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier.d.ts}

@@ -1,5 +1,3 @@
-import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
-
 /**
  * Multi-layer verifier — ordered pipeline of verification layers.
  *
@@ -23,10 +21,10 @@ import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
  * Both primitives compose — JudgeRunner can be invoked as a single
  * layer inside a MultiLayerVerifier if that suits the caller.
  */
-
-type LayerStatus = 'pass' | 'fail' | 'skipped' | 'error' | 'timeout';
-type Severity = 'critical' | 'major' | 'minor' | 'info';
-interface Finding {
+import type { DefaultVerdict } from './verdict';
+export type LayerStatus = 'pass' | 'fail' | 'skipped' | 'error' | 'timeout';
+export type Severity = 'critical' | 'major' | 'minor' | 'info';
+export interface Finding {
     severity: Severity;
     message: string;
     evidence?: string;
@@ -39,7 +37,7 @@ interface Finding {
      */
     detail?: Record<string, unknown>;
 }
-interface LayerResult {
+export interface LayerResult {
     layer: string;
     status: LayerStatus;
     /** 0..1 score, optional — layers that don't produce a numeric score omit. */
@@ -60,7 +58,7 @@ interface LayerResult {
     /** Any rich per-layer detail — rendered as-is by consumers that know the layer. */
     detail?: Record<string, unknown>;
 }
-interface VerifyContext<Env = unknown> {
+export interface VerifyContext<Env = unknown> {
     /** Per-run opaque context the caller provides. Layers destructure what they need. */
     env: Env;
     /** Previously-computed results from layers that already ran. */
@@ -68,7 +66,7 @@ interface VerifyContext<Env = unknown> {
     /** Signal — if aborted, layers MUST bail within reasonable wall. */
     signal: AbortSignal;
 }
-interface Layer<Env = unknown> {
+export interface Layer<Env = unknown> {
     name: string;
     /** Stages that must have `status: 'pass'` before this layer runs. */
     dependsOn?: string[];
@@ -87,7 +85,7 @@ interface Layer<Env = unknown> {
     capMs?: number;
     run: (ctx: VerifyContext<Env>) => Promise<LayerResult> | LayerResult;
 }
-interface VerifyOptions<Env = unknown> {
+export interface VerifyOptions<Env = unknown> {
     env: Env;
     /**
      * Overall wall cap. Default: sum of layer capMs, or Infinity if any layer
@@ -100,7 +98,7 @@ interface VerifyOptions<Env = unknown> {
 /** Extends the substrate verdict spine: `valid` = `allPass` and `score` =
  *  `blendedScore` — derived where the report is aggregated, so spine
  *  consumers (drivers, gates) read this report without an adapter. */
-interface VerificationReport extends DefaultVerdict {
+export interface VerificationReport extends DefaultVerdict {
     layers: LayerResult[];
     passCount: number;
     failCount: number;
@@ -128,7 +126,7 @@ interface VerificationReport extends DefaultVerdict {
  * overall score being >= 0.7. Now we trust the judge's own `severity` field:
  * `critical` findings veto; `major`/`minor` reduce the score but don't veto.
  */
-declare function gradeSemanticStatus(input: {
+export declare function gradeSemanticStatus(input: {
     score: number;
     findings: Array<{
         severity: Severity;
@@ -138,10 +136,9 @@ declare function gradeSemanticStatus(input: {
     available: boolean;
     threshold?: number;
 }): LayerStatus;
-declare class MultiLayerVerifier<Env = unknown> {
+export declare class MultiLayerVerifier<Env = unknown> {
     private readonly layers;
     constructor(layers: Layer<Env>[]);
     run(opts: VerifyOptions<Env>): Promise<VerificationReport>;
 }
-
-export { type Finding as F, type LayerResult as L, MultiLayerVerifier as M, type Severity as S, type VerifyOptions as V, type VerificationReport as a, type Layer as b, type VerifyContext as c, type LayerStatus as d, gradeSemanticStatus as g };
+//# sourceMappingURL=multi-layer-verifier.d.ts.map

package/dist/multi-layer-verifier.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-layer-verifier.d.ts","sourceRoot":"","sources":["../src/multi-layer-verifier.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,WAAW,CAAA;AAI/C,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAA;AAE3E,MAAM,MAAM,QAAQ,GAAG,UAAU,GAAG,OAAO,GAAG,OAAO,GAAG,MAAM,CAAA;AAE9D,MAAM,WAAW,OAAO;IACtB,QAAQ,EAAE,QAAQ,CAAA;IAClB,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,mFAAmF;IACnF,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;;;OAIG;IACH,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACjC;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,WAAW,CAAA;IACnB,6EAA6E;IAC7E,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,OAAO,EAAE,CAAA;IACnB,+CAA+C;IAC/C,MAAM,CAAC,EAAE,MAAM,CAAA;IACf;;;;;;;OAOG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,CAAA;IAC3C,mFAAmF;IACnF,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACjC;AAED,MAAM,WAAW,aAAa,CAAC,GAAG,GAAG,OAAO;IAC1C,qFAAqF;IACrF,GAAG,EAAE,GAAG,CAAA;IACR,gEAAgE;IAChE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAA;IAClC,oEAAoE;IACpE,MAAM,EAAE,WAAW,CAAA;CACpB;AAED,MAAM,WAAW,KAAK,CAAC,GAAG,GAAG,OAAO;IAClC,IAAI,EAAE,MAAM,CAAA;IACZ,qEAAqE;IACrE,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;IACpB;;;OAGG;IACH,MAAM,CAAC,EAAE,MAAM,CAAA;IACf;;;;OAIG;IACH,sBAAsB,CAAC,EAAE,OAAO,CAAA;IAChC,gFAAgF;IAChF,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,GAAG,EAAE,CAAC,GAAG,EAAE,aAAa,CAAC,GAAG,CAAC,KAAK,OAAO,CAAC,WAAW,CAAC,GAAG,WAAW,CAAA;CACrE;AAED,MAAM,WAAW,aAAa,CAAC,GAAG,GAAG,OAAO;IAC1C,GAAG,EAAE,GAAG,CAAA;IACR;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,qDAAqD;IACrD,OAAO,CAAC,EAAE,CAAC,MAAM,EAAE,WAAW,KAAK,IAAI,CAAA;CACxC;AAED;;sEAEsE;AACtE,MAAM,WAAW,kBAAmB,SAAQ,cAAc;IACxD,MAAM,EAAE,WAAW,EAAE,CAAA;IACrB,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,4EAA4E;IAC5E,OAAO,EAAE,OAAO,CAAA;IAChB;;;OAGG;IACH,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;CACnB;AAID;;;;;;;;;;GAUG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE;IACzC,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,KAAK,CAAC;QAAE,QAAQ,EAAE,QAAQ,CAAC;QAAC,OAAO,CAAC,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC1E,SAAS,EAAE,OAAO,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB,GAAG,WAAW,CAOd;AAID,qBAAa,kBAAkB,CAAC,GAAG,GAAG,OAAO;IAC/B,OAAO,CAAC,QAAQ,CAAC,MAAM;gBAAN,MAAM,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE;IAiB3C,GAAG,CAAC,IAAI,EAAE,aAAa,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,kBAAkB,CAAC;CAgFjE"}

package/dist/multi-layer-verifier.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=multi-layer-verifier.test.d.ts.map

package/dist/multi-layer-verifier.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-layer-verifier.test.d.ts","sourceRoot":"","sources":["../src/multi-layer-verifier.test.ts"],"names":[],"mappings":""}

package/dist/multi-toolchain-layer.d.ts

@@ -0,0 +1,80 @@
+/**
+ * Multi-toolchain layer factory + merge helper.
+ *
+ * Some verification stages (install, typecheck, build, lint) run the
+ * SAME logical layer across multiple parallel adapters — pnpm AND npm
+ * AND cargo AND forge for a polyglot scaffold. The verifier presents
+ * one row per stage; the toolchain breakdown lives in `findings.detail`.
+ *
+ * This module provides the merge: take N independent `LayerResult`s
+ * (one per adapter) and reduce them to a single `LayerResult` whose
+ * status is the worst of the parts and whose findings cite the adapter
+ * that produced each one. Plus a {@link multiToolchainLayer} factory
+ * that runs the adapter calls in parallel + applies the reducer.
+ *
+ * Pure utility — composes with {@link MultiLayerVerifier}.{run}.
+ */
+import type { Layer, LayerResult, VerifyContext } from './multi-layer-verifier';
+export interface AdapterRun {
+    /** Identifier for the adapter (e.g. 'pnpm', 'npm', 'cargo', 'forge'). */
+    adapter: string;
+    result: LayerResult;
+}
+export interface MergeOptions {
+    /**
+     * How to combine per-adapter `durationMs`. Default `'max'` (parallel
+     * wall-clock). Set `'sum'` when reporting total work done across
+     * adapters rather than wall time.
+     */
+    mergeDuration?: 'max' | 'sum';
+    /**
+     * Prefix finding messages with a per-adapter tag (e.g. `[pnpm] typecheck failed`).
+     * Default: no prefix (renderers read `detail.adapter` instead).
+     */
+    messagePrefixer?: (adapter: string) => string;
+    /**
+     * How to reduce per-adapter `LayerResult.diagnostics` into the merged
+     * result's diagnostics. `'max'` (default) — for each key, merged =
+     * max across adapters where value is non-null (matches "if ANY adapter
+     * saw N errors, merged saw N"). `'sum'` — sum non-null values.
+     */
+    mergeDiagnostics?: 'max' | 'sum';
+}
+/**
+ * Reduce N adapter runs to a single `LayerResult` for a logical layer.
+ *
+ *   - status: worst of the parts (pass < skipped < fail < timeout < error)
+ *   - score: weighted mean of numeric scores (skip = no contribution)
+ *   - findings: union, each tagged with `detail.adapter`
+ *   - durationMs: `mergeDuration` option (default 'max' for parallel wall-clock)
+ *   - diagnostics: `mergeDiagnostics` option (default 'max' per key)
+ *   - reason: " · "-joined `name: status` per adapter
+ */
+export declare function mergeLayerResults(name: string, perAdapter: AdapterRun[], options?: MergeOptions): LayerResult;
+export interface MultiToolchainLayerConfig<Env, Adapter> {
+    name: string;
+    adapters: ReadonlyArray<Adapter>;
+    /** Adapter identifier — used in findings + reason. */
+    adapterName: (a: Adapter) => string;
+    /** Run a single adapter against the verify context. */
+    run: (a: Adapter, ctx: VerifyContext<Env>) => Promise<LayerResult> | LayerResult;
+    dependsOn?: string[];
+    weight?: number;
+    failContributesToScore?: boolean;
+    capMs?: number;
+    /**
+     * Per-adapter parallel cap. Defaults to 8 — defense in depth against a
+     * caller passing 50 adapters and fanning out 50 simultaneous subprocesses.
+     * Adapters that need higher concurrency raise this explicitly.
+     */
+    maxParallel?: number;
+}
+/**
+ * Build a {@link Layer} that fans the same logical stage across N adapters
+ * in parallel and merges via {@link mergeLayerResults}.
+ *
+ * Per-adapter throws are caught + converted to `status: 'error'` results
+ * so one bad adapter doesn't poison the whole layer.
+ */
+export declare function multiToolchainLayer<Env, Adapter>(config: MultiToolchainLayerConfig<Env, Adapter>): Layer<Env>;
+//# sourceMappingURL=multi-toolchain-layer.d.ts.map

package/dist/multi-toolchain-layer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-toolchain-layer.d.ts","sourceRoot":"","sources":["../src/multi-toolchain-layer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,KAAK,EACV,KAAK,EACL,WAAW,EAGX,aAAa,EACd,MAAM,wBAAwB,CAAA;AAiC/B,MAAM,WAAW,UAAU;IACzB,yEAAyE;IACzE,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,WAAW,CAAA;CACpB;AAED,MAAM,WAAW,YAAY;IAC3B;;;;OAIG;IACH,aAAa,CAAC,EAAE,KAAK,GAAG,KAAK,CAAA;IAC7B;;;OAGG;IACH,eAAe,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAA;IAC7C;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,KAAK,GAAG,KAAK,CAAA;CACjC;AAED;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,MAAM,EACZ,UAAU,EAAE,UAAU,EAAE,EACxB,OAAO,GAAE,YAAiB,GACzB,WAAW,CAiFb;AAID,MAAM,WAAW,yBAAyB,CAAC,GAAG,EAAE,OAAO;IACrD,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,EAAE,aAAa,CAAC,OAAO,CAAC,CAAA;IAChC,sDAAsD;IACtD,WAAW,EAAE,CAAC,CAAC,EAAE,OAAO,KAAK,MAAM,CAAA;IACnC,uDAAuD;IACvD,GAAG,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,GAAG,EAAE,aAAa,CAAC,GAAG,CAAC,KAAK,OAAO,CAAC,WAAW,CAAC,GAAG,WAAW,CAAA;IAChF,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;IACpB,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,sBAAsB,CAAC,EAAE,OAAO,CAAA;IAChC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;;;OAIG;IACH,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,EAC9C,MAAM,EAAE,yBAAyB,CAAC,GAAG,EAAE,OAAO,CAAC,GAC9C,KAAK,CAAC,GAAG,CAAC,CAuDZ"}

package/dist/multi-toolchain-layer.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=multi-toolchain-layer.test.d.ts.map

package/dist/multi-toolchain-layer.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-toolchain-layer.test.d.ts","sourceRoot":"","sources":["../src/multi-toolchain-layer.test.ts"],"names":[],"mappings":""}

package/dist/multishot/default-tools.d.ts

@@ -0,0 +1,34 @@
+import type { MultishotToolDefinition, MultishotToolExecutor } from './types';
+export declare const DEFAULT_RESEARCHER_MODEL = "openai/gpt-4o-mini";
+export declare const DEFAULT_CODER_MODEL = "openai/gpt-4o-mini";
+export interface DefaultResearcherConfig {
+    /** Replace the system prompt to bias the researcher toward a domain's
+     *  citation style. Defaults to a generic "cite sources by name" prompt. */
+    systemPrompt?: string;
+    model?: string;
+}
+export interface DefaultCoderConfig {
+    /** Replace the system prompt to bias the coder toward a language /
+     *  framework / artifact style. */
+    systemPrompt?: string;
+    model?: string;
+}
+export declare const DEFAULT_DELEGATE_RESEARCH_TOOL: MultishotToolDefinition;
+export declare const DEFAULT_DELEGATE_CODE_TOOL: MultishotToolDefinition;
+export declare function createResearchExecutor(config?: DefaultResearcherConfig): MultishotToolExecutor;
+export declare function createCodeExecutor(config?: DefaultCoderConfig): MultishotToolExecutor;
+export interface DefaultToolsConfig {
+    research?: DefaultResearcherConfig;
+    code?: DefaultCoderConfig;
+    /** When true (default), each tool result is recorded as a typed artifact:
+     *  research → type='research', code → type='code'. */
+    recordArtifacts?: boolean;
+}
+export interface DefaultToolsBundle {
+    tools: MultishotToolDefinition[];
+    executors: Record<string, MultishotToolExecutor>;
+    artifactTypeFor: (toolName: string) => string | undefined;
+}
+export declare function defaultDelegationTools(config?: DefaultToolsConfig): DefaultToolsBundle;
+export { defaultRouterBaseUrl } from './router';
+//# sourceMappingURL=default-tools.d.ts.map

package/dist/multishot/default-tools.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"default-tools.d.ts","sourceRoot":"","sources":["../../src/multishot/default-tools.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,uBAAuB,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAA;AAE7E,eAAO,MAAM,wBAAwB,uBAAuB,CAAA;AAC5D,eAAO,MAAM,mBAAmB,uBAAuB,CAAA;AAEvD,MAAM,WAAW,uBAAuB;IACtC;+EAC2E;IAC3E,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,kBAAkB;IACjC;sCACkC;IAClC,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAQD,eAAO,MAAM,8BAA8B,EAAE,uBAkB5C,CAAA;AAED,eAAO,MAAM,0BAA0B,EAAE,uBAkBxC,CAAA;AAED,wBAAgB,sBAAsB,CACpC,MAAM,GAAE,uBAA4B,GACnC,qBAAqB,CAoBvB;AAED,wBAAgB,kBAAkB,CAAC,MAAM,GAAE,kBAAuB,GAAG,qBAAqB,CAoBzF;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,EAAE,uBAAuB,CAAA;IAClC,IAAI,CAAC,EAAE,kBAAkB,CAAA;IACzB;0DACsD;IACtD,eAAe,CAAC,EAAE,OAAO,CAAA;CAC1B;AAED,MAAM,WAAW,kBAAkB;IACjC,KAAK,EAAE,uBAAuB,EAAE,CAAA;IAChC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAA;IAChD,eAAe,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,MAAM,GAAG,SAAS,CAAA;CAC1D;AAED,wBAAgB,sBAAsB,CAAC,MAAM,GAAE,kBAAuB,GAAG,kBAAkB,CAU1F;AAED,OAAO,EAAE,oBAAoB,EAAE,MAAM,UAAU,CAAA"}

package/dist/multishot/index.d.ts

@@ -1,290 +1,7 @@
-import { J as JudgeScore } from '../types-BU-7W85F.js';
-import { AgentProfile } from '@tangle-network/sandbox';
-import { M as MatrixResult } from '../types-mn5Aqk7x.js';
-import '../run-record-e7vj1uZQ.js';
-import '../errors-CzMUYo7b.js';
-import '../schema-m0gsnbt3.js';
-import '../verdict-C9MlYujm.js';
-
-interface MultishotMessage {
-    role: 'user' | 'assistant' | 'tool';
-    content: string;
-    toolCallId?: string;
-    toolCalls?: Array<{
-        id: string;
-        name: string;
-        args: Record<string, unknown>;
-    }>;
-}
-interface MultishotArtifact {
-    type: string;
-    turn: number;
-    invocation: {
-        name: string;
-        args: Record<string, unknown>;
-    };
-    content: string;
-}
-interface MultishotResult {
-    transcript: MultishotMessage[];
-    artifacts: MultishotArtifact[];
-    toolCalls: number;
-    durationMs: number;
-    costUsd: number;
-}
-interface MultishotToolDefinition {
-    type: 'function';
-    function: {
-        name: string;
-        description: string;
-        parameters: Record<string, unknown>;
-    };
-}
-type MultishotToolExecutor = (args: Record<string, unknown>, ctx: {
-    apiKey: string;
-    baseUrl: string;
-    signal?: AbortSignal;
-}) => Promise<{
-    content: string;
-    costUsd: number;
-}>;
-interface MultishotPersona {
-    /** Stable identifier — used for per-cell artifact paths + matrix axis keys. */
-    id: string;
-    /** Per-domain payload (income/profile/voice/etc.) shaped by the consumer. */
-    [k: string]: unknown;
-}
-interface MultishotShape<TPersona extends MultishotPersona> {
-    /** Opening user message (turn 0) — the persona's first ask. */
-    buildOpener: (persona: TPersona) => string;
-    /** System prompt the driver LLM uses to roleplay the persona. Should set
-     *  voice, goals, constraints, time-pressure, and the "never go silent" rule. */
-    buildDriverSystemPrompt: (persona: TPersona) => string;
-}
-declare class MultishotDriverEmptyError extends Error {
-    readonly turn: number;
-    constructor(turn: number);
-}
-
-interface RouterCompletionRequest {
-    apiKey: string;
-    baseUrl: string;
-    model: string;
-    messages: Array<Record<string, unknown>>;
-    tools?: MultishotToolDefinition[];
-    temperature?: number;
-    maxTokens?: number;
-    signal?: AbortSignal;
-}
-interface RouterToolCall {
-    id: string;
-    type: 'function';
-    function: {
-        name: string;
-        arguments: string;
-    };
-}
-interface RouterCompletionResponse {
-    message: {
-        content?: string | null;
-        tool_calls?: RouterToolCall[];
-    };
-    usage?: {
-        prompt_tokens?: number;
-        completion_tokens?: number;
-    };
-}
-declare function routerCompletion(req: RouterCompletionRequest): Promise<RouterCompletionResponse>;
-declare function estimateRouterCost(model: string, usage?: {
-    prompt_tokens?: number;
-    completion_tokens?: number;
-}): number;
-declare function defaultRouterBaseUrl(): string;
-declare function requireRouterApiKey(): string;
-
-declare const DEFAULT_RESEARCHER_MODEL = "openai/gpt-4o-mini";
-declare const DEFAULT_CODER_MODEL = "openai/gpt-4o-mini";
-interface DefaultResearcherConfig {
-    /** Replace the system prompt to bias the researcher toward a domain's
-     *  citation style. Defaults to a generic "cite sources by name" prompt. */
-    systemPrompt?: string;
-    model?: string;
-}
-interface DefaultCoderConfig {
-    /** Replace the system prompt to bias the coder toward a language /
-     *  framework / artifact style. */
-    systemPrompt?: string;
-    model?: string;
-}
-declare const DEFAULT_DELEGATE_RESEARCH_TOOL: MultishotToolDefinition;
-declare const DEFAULT_DELEGATE_CODE_TOOL: MultishotToolDefinition;
-declare function createResearchExecutor(config?: DefaultResearcherConfig): MultishotToolExecutor;
-declare function createCodeExecutor(config?: DefaultCoderConfig): MultishotToolExecutor;
-interface DefaultToolsConfig {
-    research?: DefaultResearcherConfig;
-    code?: DefaultCoderConfig;
-    /** When true (default), each tool result is recorded as a typed artifact:
-     *  research → type='research', code → type='code'. */
-    recordArtifacts?: boolean;
-}
-interface DefaultToolsBundle {
-    tools: MultishotToolDefinition[];
-    executors: Record<string, MultishotToolExecutor>;
-    artifactTypeFor: (toolName: string) => string | undefined;
-}
-declare function defaultDelegationTools(config?: DefaultToolsConfig): DefaultToolsBundle;
-
-declare const DEFAULT_JUDGE_MODEL = "openai/gpt-4o-mini";
-interface JudgeDimension {
-    /** JSON field name + score key. */
-    key: string;
-    /** Description shown in the judge's user prompt. */
-    description: string;
-}
-interface JudgeConfig<TInput> {
-    /** Display name (for trace + log). */
-    name: string;
-    /** Model used for this judge. */
-    model?: string;
-    /** 0-10 scored dimensions. */
-    dimensions: JudgeDimension[];
-    /** Judge system prompt — sets persona + JSON-only constraint. */
-    systemPrompt: string;
-    /** Build the user prompt from the typed input. Must include "Respond with
-     *  ONLY this JSON: { ... }" listing each dimension key. */
-    buildPrompt: (input: TInput) => string;
-    /** Optional model + api overrides. */
-    apiKey?: string;
-    baseUrl?: string;
-}
-declare function runJudge<TInput>(judge: JudgeConfig<TInput>, input: TInput): Promise<JudgeScore>;
-/** Convenience: stringified dimension list for inclusion in a judge prompt.
- *  Returns lines like `- audience_fit: Does this match what the audience cares about? (0-10)`. */
-declare function renderDimensions(dims: readonly JudgeDimension[]): string;
-/** Convenience: build the "Respond with ONLY this JSON" footer for a judge prompt. */
-declare function renderJsonFooter(dims: readonly JudgeDimension[]): string;
-
-interface ConversationJudgeInput<TPersona extends MultishotPersona> {
-    transcript: MultishotMessage[];
-    persona: TPersona;
-}
-interface ArtifactJudgeInput<TPersona extends MultishotPersona> {
-    artifact: MultishotArtifact;
-    persona: TPersona;
-}
-interface MultishotJudges<TPersona extends MultishotPersona> {
-    /** Scores the full transcript end-to-end (always runs). */
-    conversation: JudgeConfig<ConversationJudgeInput<TPersona>>;
-    /** Scores each code-type artifact. Optional — omit when domain has no code artifacts. */
-    codeReview?: JudgeConfig<ArtifactJudgeInput<TPersona>>;
-    /** Scores each non-code (research/content/template) artifact. Optional. */
-    contentQuality?: JudgeConfig<ArtifactJudgeInput<TPersona>>;
-    /** Which artifact types route to codeReview. Defaults to ['code']. */
-    codeArtifactTypes?: string[];
-    /** Which artifact types route to contentQuality. Defaults to ['research']. */
-    contentArtifactTypes?: string[];
-}
-interface CellCompositeScore {
-    composite: number;
-    conversation: JudgeScore;
-    codeReview?: {
-        perArtifact: Array<JudgeScore & {
-            turn: number;
-            type: string;
-        }>;
-        composite: number;
-    };
-    contentQuality?: {
-        perArtifact: Array<JudgeScore & {
-            turn: number;
-            type: string;
-        }>;
-        composite: number;
-    };
-}
-interface RunMultishotMatrixOptions<TPersona extends MultishotPersona> {
-    /** AgentProfile axis (matrix primary). */
-    profiles: Array<{
-        id: string;
-        value: AgentProfile;
-    }>;
-    /** Persona axis. */
-    personas: TPersona[];
-    /** Persona-shaping callbacks. */
-    shape: MultishotShape<TPersona>;
-    /** Judge configurations. */
-    judges: MultishotJudges<TPersona>;
-    /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */
-    tools?: MultishotToolDefinition[];
-    /** Map from tool name → inline executor. Must align with `tools`. */
-    toolExecutors?: Record<string, MultishotToolExecutor>;
-    /** Tool name → artifact type label. Defaults to research/code mapping. */
-    artifactTypeFor?: (toolName: string) => string | undefined;
-    /** Where per-cell artifacts land. Cells write to `<runDir>/<profileId>/<personaId>/rep-N/`. */
-    runDir: string;
-    /** Replicates per (profile, persona) cell. */
-    reps?: number;
-    /** Max conversation turns per cell. */
-    maxTurns?: number;
-    /** Max concurrent cells. */
-    maxConcurrency?: number;
-    /** Total $ ceiling across the matrix; cells aborted past this. */
-    costCeiling?: number;
-    /** Agent model. */
-    agentModel?: string;
-    /** Driver model. */
-    driverModel?: string;
-    /** Pass-thru fields. */
-    apiKey?: string;
-    baseUrl?: string;
-}
-interface CellOutput {
-    turns: number;
-    toolCalls: number;
-    artifactCount: number;
-}
-interface CellCompositeInput {
-    conversation: JudgeScore;
-    /** Present iff the codeReview judge is configured. */
-    codeReviews?: ReadonlyArray<JudgeScore>;
-    /** Present iff the contentQuality judge is configured. */
-    contentReviews?: ReadonlyArray<JudgeScore>;
-}
-/** Cell composite = mean over configured judge slots, excluding failed
- *  scores: a failed conversation judge or an all-failed artifact slot carries
- *  no signal and is dropped from the mean. `composite` is 0 only when EVERY
- *  configured slot failed (`allJudgesFailed` distinguishes that from a real
- *  zero). Pure — exported for deterministic testing. */
-declare function computeCellComposite(input: CellCompositeInput): {
-    composite: number;
-    codeComposite: number;
-    contentComposite: number;
-    allJudgesFailed: boolean;
-};
-interface RunMultishotMatrixResult {
-    matrix: MatrixResult<CellOutput>;
-}
-declare function runMultishotMatrix<TPersona extends MultishotPersona>(opts: RunMultishotMatrixOptions<TPersona>): Promise<RunMultishotMatrixResult>;
-
-interface RunMultishotOptions<TPersona extends MultishotPersona> {
-    profile: AgentProfile;
-    persona: TPersona;
-    shape: MultishotShape<TPersona>;
-    /** Tool definitions advertised to the agent. Defaults to delegate_research + delegate_code. */
-    tools?: MultishotToolDefinition[];
-    /** Map from tool name → executor invoked inline when the agent emits a tool_call. */
-    toolExecutors?: Record<string, MultishotToolExecutor>;
-    /** Map from tool name → artifact type label written into MultishotArtifact.type.
-     *  Tools without a mapping still execute, but their results aren't surfaced as
-     *  typed artifacts (only as tool messages in the transcript). */
-    artifactTypeFor?: (toolName: string) => string | undefined;
-    maxTurns?: number;
-    agentModel?: string;
-    driverModel?: string;
-    apiKey?: string;
-    baseUrl?: string;
-    signal?: AbortSignal;
-}
-declare function runMultishot<TPersona extends MultishotPersona>(opts: RunMultishotOptions<TPersona>): Promise<MultishotResult>;
-
-export { type ArtifactJudgeInput, type CellCompositeInput, type CellCompositeScore, type ConversationJudgeInput, DEFAULT_CODER_MODEL, DEFAULT_DELEGATE_CODE_TOOL, DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_JUDGE_MODEL, DEFAULT_RESEARCHER_MODEL, type DefaultCoderConfig, type DefaultResearcherConfig, type DefaultToolsBundle, type DefaultToolsConfig, type JudgeConfig, type JudgeDimension, JudgeScore, type MultishotArtifact, MultishotDriverEmptyError, type MultishotJudges, type MultishotMessage, type MultishotPersona, type MultishotResult, type MultishotShape, type MultishotToolDefinition, type MultishotToolExecutor, type RouterCompletionRequest, type RouterCompletionResponse, type RouterToolCall, type RunMultishotMatrixOptions, type RunMultishotMatrixResult, type RunMultishotOptions, computeCellComposite, createCodeExecutor, createResearchExecutor, defaultDelegationTools, defaultRouterBaseUrl, estimateRouterCost, renderDimensions, renderJsonFooter, requireRouterApiKey, routerCompletion, runJudge, runMultishot, runMultishotMatrix };
+export { createCodeExecutor, createResearchExecutor, DEFAULT_CODER_MODEL, DEFAULT_DELEGATE_CODE_TOOL, DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_RESEARCHER_MODEL, type DefaultCoderConfig, type DefaultResearcherConfig, type DefaultToolsBundle, type DefaultToolsConfig, defaultDelegationTools, } from './default-tools';
+export { DEFAULT_JUDGE_MODEL, type JudgeConfig, type JudgeDimension, type JudgeScore, renderDimensions, renderJsonFooter, runJudge, } from './judges';
+export { type ArtifactJudgeInput, type CellCompositeInput, type CellCompositeScore, type ConversationJudgeInput, computeCellComposite, type MultishotJudges, type RunMultishotMatrixOptions, type RunMultishotMatrixResult, runMultishotMatrix, } from './matrix';
+export { type RunMultishotOptions, runMultishot } from './multishot';
+export { defaultRouterBaseUrl, estimateRouterCost, type RouterCompletionRequest, type RouterCompletionResponse, type RouterToolCall, requireRouterApiKey, routerCompletion, } from './router';
+export { type MultishotArtifact, MultishotDriverEmptyError, type MultishotMessage, type MultishotPersona, type MultishotResult, type MultishotShape, type MultishotToolDefinition, type MultishotToolExecutor, } from './types';
+//# sourceMappingURL=index.d.ts.map

package/dist/multishot/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/multishot/index.ts"],"names":[],"mappings":"AAEA,OAAO,EACL,kBAAkB,EAClB,sBAAsB,EACtB,mBAAmB,EACnB,0BAA0B,EAC1B,8BAA8B,EAC9B,wBAAwB,EACxB,KAAK,kBAAkB,EACvB,KAAK,uBAAuB,EAC5B,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,sBAAsB,GACvB,MAAM,iBAAiB,CAAA;AAExB,OAAO,EACL,mBAAmB,EACnB,KAAK,WAAW,EAChB,KAAK,cAAc,EACnB,KAAK,UAAU,EACf,gBAAgB,EAChB,gBAAgB,EAChB,QAAQ,GACT,MAAM,UAAU,CAAA;AACjB,OAAO,EACL,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,KAAK,sBAAsB,EAC3B,oBAAoB,EACpB,KAAK,eAAe,EACpB,KAAK,yBAAyB,EAC9B,KAAK,wBAAwB,EAC7B,kBAAkB,GACnB,MAAM,UAAU,CAAA;AACjB,OAAO,EAAE,KAAK,mBAAmB,EAAE,YAAY,EAAE,MAAM,aAAa,CAAA;AAEpE,OAAO,EACL,oBAAoB,EACpB,kBAAkB,EAClB,KAAK,uBAAuB,EAC5B,KAAK,wBAAwB,EAC7B,KAAK,cAAc,EACnB,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,UAAU,CAAA;AAEjB,OAAO,EACL,KAAK,iBAAiB,EACtB,yBAAyB,EACzB,KAAK,gBAAgB,EACrB,KAAK,gBAAgB,EACrB,KAAK,eAAe,EACpB,KAAK,cAAc,EACnB,KAAK,uBAAuB,EAC5B,KAAK,qBAAqB,GAC3B,MAAM,SAAS,CAAA"}