@principles/core 1.151.0 → 1.153.0

package/dist/runtime-v2/internalization/evaluator-runner.d.ts

@@ -30,11 +30,20 @@ import type { TaskRecord } from '../task-status.js';
 import { type PDErrorCategory } from '../error-categories.js';
 import { BasePeerRunner } from '../runner/base-peer-runner.js';
 import type { PeerRunnerOptions, PeerRunnerDeps, PeerRunnerResult, PeerRunnerValidationResult } from '../runner/peer-runner-types.js';
+import { type RefinerRuleHostGateDeps } from './refiner-rulehost-gate.js';
 /** Context built by EvaluatorRunner.buildContext() and consumed by invokeRuntime(). */
 interface EvaluatorContext {
     readonly contextHash: string;
     readonly artificerArtifact: string | null;
     readonly sourceArtificerArtifactId: string | null;
+    /**
+     * Scribe principle artifact contentJson (RuleHost MVP Activation, PRD Decision 12).
+     * Loaded so the evaluator LLM can judge code intentConsistency/scopePrecision
+     * against the original principle text. Null when no scribe artifact is resolvable
+     * (V1 artificer output, or scribeArtifactId missing/malformed).
+     */
+    readonly scribeArtifact: string | null;
+    readonly sourceScribeArtifactId: string | null;
 }
 export type EvaluatorRunnerResultStatus = 'succeeded' | 'failed' | 'retried';
 export interface EvaluatorRunnerResult {
@@ -49,7 +58,19 @@ export interface EvaluatorRunnerResult {
     readonly failureReason?: string;
     readonly attemptCount: number;
 }
-export type EvaluatorRunnerOptions = PeerRunnerOptions;
+/**
+ * EvaluatorRunner options. Extends PeerRunnerOptions with an optional
+ * RuleHost sandbox gate deps (PRI-426). When `gateDeps` is provided AND the
+ * evaluator output is V2 (code-bearing), succeedTask runs a single-round
+ * adversarial sandbox replay and populates `adversarialResult`.
+ *
+ * When `gateDeps` is absent, V2 outputs still validate but no replay runs —
+ * this preserves backward compatibility for callers not yet wired to the
+ * sandbox (e.g. V1-only test fixtures, pre-Phase-6 assembly).
+ */
+export interface EvaluatorRunnerOptions extends PeerRunnerOptions {
+    readonly gateDeps?: RefinerRuleHostGateDeps;
+}
 export interface ResolvedEvaluatorRunnerOptions {
     readonly pollIntervalMs: number;
     readonly timeoutMs: number;
@@ -65,7 +86,13 @@ export interface EvaluatorRunnerDeps extends PeerRunnerDeps {
 }
 export declare class EvaluatorRunner extends BasePeerRunner<EvaluatorContext, EvaluatorOutputV1> {
     private readonly validator;
-    constructor(deps: EvaluatorRunnerDeps, options: PeerRunnerOptions);
+    /**
+     * Optional RuleHost sandbox gate deps (PRI-426). When present and the output
+     * is V2, succeedTask runs a single adversarial sandbox replay. Absent = no
+     * replay (backward compatible).
+     */
+    private readonly gateDeps;
+    constructor(deps: EvaluatorRunnerDeps, options: EvaluatorRunnerOptions);
     get permanentErrorCategories(): ReadonlySet<PDErrorCategory>;
     buildContext(taskId: string): Promise<EvaluatorContext>;
     invokeRuntime(taskId: string, context: EvaluatorContext): Promise<RunHandle>;
@@ -86,6 +113,59 @@ export declare class EvaluatorRunner extends BasePeerRunner<EvaluatorContext, Ev
      * Validates sourceTrace.scribeArtifactId consistency (ERR-004, ERR-008).
      */
     protected checkLineageIntegrity(taskId: string, output: EvaluatorOutputV1, _context: EvaluatorContext): void;
+    /**
+     * Run a single-round adversarial sandbox replay on the evaluator's V2 output
+     * (PRD Decision 11d). Pure orchestration of pure functions:
+     *   1. Skip if passive review failed (decision !== 'approved' is the LLM's
+     *      short-circuit signal — no code to defend).
+     *   2. Skip if no adversarialCases present (V2 with codeReview only).
+     *   3. Convert adversarialCases → GoldenTrace (all negative, PRI-423).
+     *   4. Merge ≥1 positive case from the Artificer golden trace. If the
+     *      artificer artifact has no goldenTraceCases (V1 mismatch), degrade:
+     *      skip replay with telemetry — do NOT crash.
+     *   5. Invoke evaluateRefinerRuleHostGate via injected gateDeps.
+     *   6. Populate adversarialResult from the gate result.
+     *
+     * Never throws — all failure modes degrade to a returned result with a
+     * structured reason (ERR-018). The caller persists the updated output.
+     */
+    private runAdversarialReplay;
+    /**
+     * Parse the Artificer artifact contentJson defensively (Runtime Contract
+     * Rule 1/2/5). Returns null on any structural issue — the caller degrades.
+     */
+    private parseArtificerArtifact;
+    /**
+     * Extract structurally-valid positive GoldenTraceCases from the Artificer
+     * goldenTraceCases array. Uses buildGoldenTraceFromArtificer (which re-
+     * validates each case) and filters for kind='positive'. Returns [] when
+     * the input is missing/malformed — the caller degrades to a skip (ERR-069:
+     * never trust unvalidated candidates).
+     */
+    private extractPositiveCases;
+    /**
+     * Map sandbox failed cases to EvaluatorAdversarialResult.failedCases.
+     *
+     * The sandbox reports failedCases by caseId; we enrich each with the
+     * adversarial case's attackType and expectedDecision. Cases not found in
+     * the adversarial set (e.g. the merged positive case failed — which would
+     * indicate a code bug, not an adversarial failure) are reported with
+     * attackType='boundary' as a safe default and a note in the rationale
+     * (Runtime Contract Rule 9: graceful degradation includes a reason).
+     */
+    private mapFailedCases;
+    /**
+     * Assemble and persist the rule artifact when adversarial replay passed
+     * (PRD Decision 5). The rule artifact carries implementationCode + the
+     * Artificer full golden trace + ruleHostGateDecision, with artifactKind='rule'.
+     * After a successful write, marks the artifact 'validated' so RuleHostWriter
+     * can activate it.
+     *
+     * Returns the rule artifactId on success, null on any degradation (missing
+     * code/trace, write failure, validation-update failure). Every null path
+     * emits structured telemetry with a reason (Runtime Rule 9, ERR-018).
+     */
+    private assembleRuleArtifact;
     /**
      * Resolve the principle-bearing artifact that the evaluator should validate.
      *

package/dist/runtime-v2/internalization/evaluator-runner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"evaluator-runner.d.ts","sourceRoot":"","sources":["../../../src/runtime-v2/internalization/evaluator-runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,KAAK,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,MAAM,uBAAuB,CAAC;AACnF,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAkB,KAAK,eAAe,EAAqB,MAAM,wBAAwB,CAAC;AAIjG,OAAO,EAAE,cAAc,EAAE,MAAM,+BAA+B,CAAC;AAC/D,OAAO,KAAK,EACV,iBAAiB,EACjB,cAAc,EACd,gBAAgB,EAChB,0BAA0B,EAC3B,MAAM,gCAAgC,CAAC;AAIxC,uFAAuF;AACvF,UAAU,gBAAgB;IACxB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1C,QAAQ,CAAC,yBAAyB,EAAE,MAAM,GAAG,IAAI,CAAC;CACnD;AAID,MAAM,MAAM,2BAA2B,GAAG,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,MAAM,EAAE,2BAA2B,CAAC;IAC7C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,MAAM,CAAC,EAAE,iBAAiB,CAAC;IACpC,QAAQ,CAAC,aAAa,CAAC,EAAE,eAAe,CAAC;IACzC,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,CAAC;IAChC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAID,MAAM,MAAM,sBAAsB,GAAG,iBAAiB,CAAC;AAEvD,MAAM,WAAW,8BAA8B;IAC7C,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;IACpC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,eAAO,MAAM,gCAAgC,EAAE,QAAQ,CAAC,IAAI,CAAC,8BAA8B,EAAE,OAAO,GAAG,aAAa,CAAC,CAK3G,CAAC;AAEX,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,sBAAsB,GAAG,8BAA8B,CAS7G;AAID,MAAM,WAAW,mBAAoB,SAAQ,cAAc;IACzD,QAAQ,CAAC,SAAS,EAAE,kBAAkB,CAAC;CACxC;AAID,qBAAa,eAAgB,SAAQ,cAAc,CAAC,gBAAgB,EAAE,iBAAiB,CAAC;IACtF,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAqB;gBAEnC,IAAI,EAAE,mBAAmB,EAAE,OAAO,EAAE,iBAAiB;IAajE,IAAI,wBAAwB,IAAI,WAAW,CAAC,eAAe,CAAC,CAE3D;IAEK,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC;IA+CvD,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,SAAS,CAAC;IA4B5E,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,0BAA0B,CAAC;IA4B/G,WAAW,CACf,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,iBAAiB,EACzB,IAAI,EAAE,UAAU,EAChB,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;IAqI/C;;;;;;;OAOG;cAEgB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,GAAG,IAAI;cAUlE,oBAAoB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB,GAAG,IAAI;IAOxF;;;OAGG;cACgB,qBAAqB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB,EAAE,QAAQ,EAAE,gBAAgB,GAAG,IAAI;IAarH;;;;;;;;;;OAUG;YACW,8BAA8B;IAwD5C;;;OAGG;IAEH,OAAO,CAAC,wBAAwB;IAwBhC,OAAO,CAAC,MAAM,CAAC,QAAQ;CAGxB"}
+{"version":3,"file":"evaluator-runner.d.ts","sourceRoot":"","sources":["../../../src/runtime-v2/internalization/evaluator-runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,KAAK,EACV,iBAAiB,EAEjB,kBAAkB,EAGnB,MAAM,uBAAuB,CAAC;AAE/B,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAkB,KAAK,eAAe,EAAqB,MAAM,wBAAwB,CAAC;AAIjG,OAAO,EAAE,cAAc,EAAE,MAAM,+BAA+B,CAAC;AAC/D,OAAO,KAAK,EACV,iBAAiB,EACjB,cAAc,EACd,gBAAgB,EAChB,0BAA0B,EAC3B,MAAM,gCAAgC,CAAC;AAExC,OAAO,EAA+B,KAAK,uBAAuB,EAAE,MAAM,4BAA4B,CAAC;AAOvG,uFAAuF;AACvF,UAAU,gBAAgB;IACxB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1C,QAAQ,CAAC,yBAAyB,EAAE,MAAM,GAAG,IAAI,CAAC;IAClD;;;;;OAKG;IACH,QAAQ,CAAC,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IACvC,QAAQ,CAAC,sBAAsB,EAAE,MAAM,GAAG,IAAI,CAAC;CAChD;AAyCD,MAAM,MAAM,2BAA2B,GAAG,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,MAAM,EAAE,2BAA2B,CAAC;IAC7C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,MAAM,CAAC,EAAE,iBAAiB,CAAC;IACpC,QAAQ,CAAC,aAAa,CAAC,EAAE,eAAe,CAAC;IACzC,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,CAAC;IAChC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAID;;;;;;;;;GASG;AACH,MAAM,WAAW,sBAAuB,SAAQ,iBAAiB;IAC/D,QAAQ,CAAC,QAAQ,CAAC,EAAE,uBAAuB,CAAC;CAC7C;AAED,MAAM,WAAW,8BAA8B;IAC7C,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;IACpC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,eAAO,MAAM,gCAAgC,EAAE,QAAQ,CAAC,IAAI,CAAC,8BAA8B,EAAE,OAAO,GAAG,aAAa,CAAC,CAK3G,CAAC;AAEX,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,sBAAsB,GAAG,8BAA8B,CAS7G;AAID,MAAM,WAAW,mBAAoB,SAAQ,cAAc;IACzD,QAAQ,CAAC,SAAS,EAAE,kBAAkB,CAAC;CACxC;AAID,qBAAa,eAAgB,SAAQ,cAAc,CAAC,gBAAgB,EAAE,iBAAiB,CAAC;IACtF,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAqB;IAC/C;;;;OAIG;IACH,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAiC;gBAE9C,IAAI,EAAE,mBAAmB,EAAE,OAAO,EAAE,sBAAsB;IActE,IAAI,wBAAwB,IAAI,WAAW,CAAC,eAAe,CAAC,CAE3D;IAEK,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAwEvD,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,SAAS,CAAC;IAyC5E,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,0BAA0B,CAAC;IA4B/G,WAAW,CACf,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,iBAAiB,EACzB,IAAI,EAAE,UAAU,EAChB,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;IAqM/C;;;;;;;OAOG;cAEgB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,GAAG,IAAI;cAUlE,oBAAoB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB,GAAG,IAAI;IAOxF;;;OAGG;cACgB,qBAAqB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB,EAAE,QAAQ,EAAE,gBAAgB,GAAG,IAAI;IAarH;;;;;;;;;;;;;;;OAeG;YAEW,oBAAoB;IAgJlC;;;OAGG;IAEH,OAAO,CAAC,sBAAsB;IAiB9B;;;;;;OAMG;IAEH,OAAO,CAAC,oBAAoB;IAO5B;;;;;;;;;OASG;IAEH,OAAO,CAAC,cAAc;IAqBtB;;;;;;;;;;OAUG;YAEW,oBAAoB;IA8IlC;;;;;;;;;;OAUG;YACW,8BAA8B;IAwD5C;;;OAGG;IAEH,OAAO,CAAC,wBAAwB;IAwBhC,OAAO,CAAC,MAAM,CAAC,QAAQ;CAGxB"}