@cat-factory/orchestration 0.19.2 → 0.21.0

package/dist/modules/sandbox/SandboxService.js

@@ -0,0 +1,227 @@
+import { assertFound, ConflictError, requireWorkspace, ValidationError } from '@cat-factory/kernel';
+import { cellCount, firstVersionFromBaseline, isRunnableMatrix, listBaselines, listBuiltinFixtures, nextVersion, SANDBOX_AGENT_KINDS, baselineVersionId, sandboxKindMeta, } from '@cat-factory/sandbox';
+/** A safety ceiling on how many cells one experiment may expand to (cost guard). */
+export const MAX_SANDBOX_CELLS = 100;
+/**
+ * Load an experiment and its result grid (cells + grades). Shared by the read endpoint
+ * (`SandboxService.getExperiment`) and the run-driver's returned grid
+ * (`SandboxRunService.launch`) so the two can never compose a divergent detail shape.
+ */
+export async function composeExperimentDetail(repos, workspaceId, experimentId) {
+    const experiment = assertFound(await repos.sandboxExperimentRepository.get(workspaceId, experimentId), 'SandboxExperiment', experimentId);
+    const [runs, grades] = await Promise.all([
+        repos.sandboxRunRepository.listByExperiment(workspaceId, experimentId),
+        repos.sandboxGradeRepository.listByExperiment(workspaceId, experimentId),
+    ]);
+    return { experiment, runs, grades };
+}
+/**
+ * Management CRUD for the Sandbox (the parallel prompt/model testing surface): the
+ * shipped baselines + stored candidate prompt versions, the fixture library (builtins
+ * seeded lazily on first list, plus workspace-authored ones), and experiment definitions
+ * + their result grids. Running an experiment lives in {@link SandboxRunService}; this
+ * service is the persistence-facing half. Everything is workspace-scoped.
+ */
+export class SandboxService {
+    deps;
+    constructor(deps) {
+        this.deps = deps;
+    }
+    /** The full opt-in overview the UI loads when the Sandbox surface opens. */
+    async overview(workspaceId) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const [prompts, fixtures, experiments] = await Promise.all([
+            this.listPrompts(workspaceId),
+            this.listFixtures(workspaceId),
+            this.deps.sandboxExperimentRepository.list(workspaceId),
+        ]);
+        return {
+            agentKinds: SANDBOX_AGENT_KINDS,
+            prompts,
+            fixtures,
+            experiments,
+            maxCells: MAX_SANDBOX_CELLS,
+        };
+    }
+    // ---- prompt versions ------------------------------------------------------
+    /** The shipped baselines (synthetic) followed by stored candidate versions. */
+    async listPrompts(workspaceId, agentKind) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const baselines = listBaselines(this.deps.clock.now());
+        const candidates = agentKind
+            ? await this.deps.sandboxPromptVersionRepository.listByKind(workspaceId, agentKind)
+            : await this.deps.sandboxPromptVersionRepository.list(workspaceId);
+        const baseSlice = agentKind ? baselines.filter((b) => b.agentKind === agentKind) : baselines;
+        return [...baseSlice, ...candidates];
+    }
+    /** Clone a shipped baseline into a fresh editable candidate lineage at version 1. */
+    async clonePrompt(workspaceId, input) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const source = this.resolveBaseline(input.agentKind, input.basePromptId);
+        const meta = sandboxKindMeta(input.agentKind);
+        const name = input.name ?? meta?.label ?? input.agentKind;
+        const version = firstVersionFromBaseline({
+            agentKind: source.agentKind,
+            systemText: source.systemText,
+            basePromptId: source.basePromptId,
+        }, name, {
+            id: this.deps.idGenerator.next('sbp'),
+            createdAt: this.deps.clock.now(),
+            createdBy: null,
+            labels: input.labels ?? [],
+        });
+        await this.deps.sandboxPromptVersionRepository.upsert(workspaceId, version);
+        return version;
+    }
+    /** Append a new candidate version. The parent may be a baseline (starts a lineage) or a candidate. */
+    async saveVersion(workspaceId, input) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const fields = {
+            id: this.deps.idGenerator.next('sbp'),
+            createdAt: this.deps.clock.now(),
+            createdBy: null,
+            labels: input.labels ?? [],
+        };
+        if (input.parentId.startsWith('baseline:')) {
+            const baseline = listBaselines(this.deps.clock.now()).find((b) => b.id === input.parentId);
+            if (!baseline)
+                throw new ValidationError(`Unknown baseline prompt "${input.parentId}"`);
+            const meta = sandboxKindMeta(baseline.agentKind);
+            const version = firstVersionFromBaseline({
+                agentKind: baseline.agentKind,
+                systemText: input.systemText,
+                basePromptId: baseline.basePromptId,
+            }, meta?.label ?? baseline.agentKind, fields);
+            await this.deps.sandboxPromptVersionRepository.upsert(workspaceId, version);
+            return version;
+        }
+        const parent = assertFound(await this.deps.sandboxPromptVersionRepository.get(workspaceId, input.parentId), 'SandboxPromptVersion', input.parentId);
+        const version = nextVersion(parent, input.systemText, fields);
+        await this.deps.sandboxPromptVersionRepository.upsert(workspaceId, version);
+        return version;
+    }
+    /** Replace a candidate version's labels. */
+    async setLabels(workspaceId, id, input) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const existing = assertFound(await this.deps.sandboxPromptVersionRepository.get(workspaceId, id), 'SandboxPromptVersion', id);
+        const updated = { ...existing, labels: input.labels };
+        await this.deps.sandboxPromptVersionRepository.upsert(workspaceId, updated);
+        return updated;
+    }
+    /** Soft-archive a candidate version (hidden from the default listing). */
+    async archivePrompt(workspaceId, id) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        await this.deps.sandboxPromptVersionRepository.archive(workspaceId, id, this.deps.clock.now());
+    }
+    // ---- fixtures -------------------------------------------------------------
+    /** The fixture library, seeding the builtin fixtures on first use. */
+    async listFixtures(workspaceId) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        await this.ensureBuiltinFixtures(workspaceId);
+        return this.deps.sandboxFixtureRepository.list(workspaceId);
+    }
+    /** Create a workspace-authored fixture. */
+    async createFixture(workspaceId, input) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const fixture = {
+            id: this.deps.idGenerator.next('sbf'),
+            kind: input.kind,
+            name: input.name,
+            payload: input.payload ?? null,
+            repoRef: input.repoRef ?? null,
+            objective: input.objective ?? null,
+            origin: 'custom',
+            createdAt: this.deps.clock.now(),
+        };
+        await this.deps.sandboxFixtureRepository.upsert(workspaceId, fixture);
+        return fixture;
+    }
+    /** Remove a workspace-authored fixture. Builtin fixtures cannot be removed. */
+    async removeFixture(workspaceId, id) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const existing = await this.deps.sandboxFixtureRepository.get(workspaceId, id);
+        if (existing?.origin === 'builtin') {
+            throw new ConflictError('Builtin fixtures cannot be deleted.');
+        }
+        await this.deps.sandboxFixtureRepository.remove(workspaceId, id);
+    }
+    // ---- experiments ----------------------------------------------------------
+    async listExperiments(workspaceId) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        return this.deps.sandboxExperimentRepository.list(workspaceId);
+    }
+    /** An experiment with its result grid (cells + grades). */
+    async getExperiment(workspaceId, id) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        return composeExperimentDetail(this.deps, workspaceId, id);
+    }
+    /** Create a draft experiment. Launching it (the run-driver) expands + grades the matrix. */
+    async createExperiment(workspaceId, input) {
+        await requireWorkspace(this.deps.workspaceRepository, workspaceId);
+        const meta = sandboxKindMeta(input.agentKind);
+        if (!meta) {
+            throw new ValidationError(`"${input.agentKind}" is not a Sandbox-testable agent kind`);
+        }
+        // Refuse container kinds up front: the in-product run driver only runs inline cells,
+        // so a container experiment could be persisted but never launched. Reject at create
+        // time rather than leaving an un-launchable draft in the workspace.
+        if (meta.bucket === 'container') {
+            throw new ValidationError(`The "${input.agentKind}" agent runs in a container; container experiments are not yet supported in the Sandbox.`);
+        }
+        if (!isRunnableMatrix(input.matrix)) {
+            throw new ValidationError('The experiment matrix needs at least one prompt, model and fixture');
+        }
+        const repeats = input.repeats ?? 1;
+        const total = cellCount(input.matrix, repeats);
+        if (total > MAX_SANDBOX_CELLS) {
+            throw new ValidationError(`This matrix expands to ${total} cells; the limit is ${MAX_SANDBOX_CELLS}. Narrow the selection.`);
+        }
+        const experiment = {
+            id: this.deps.idGenerator.next('sbx'),
+            name: input.name,
+            agentKind: input.agentKind,
+            judgeModel: input.judgeModel ?? this.defaultJudgeModel(),
+            repeats,
+            status: 'draft',
+            matrix: input.matrix,
+            budgetTokens: input.budgetTokens ?? null,
+            createdAt: this.deps.clock.now(),
+            createdBy: null,
+        };
+        await this.deps.sandboxExperimentRepository.upsert(workspaceId, experiment);
+        return experiment;
+    }
+    // ---- internals ------------------------------------------------------------
+    /** Resolve the shipped baseline a clone derives from (by base-prompt id, else by kind). */
+    resolveBaseline(agentKind, basePromptId) {
+        const baselines = listBaselines(this.deps.clock.now());
+        const wantedId = basePromptId ? `baseline:${basePromptId}` : baselineVersionId(agentKind);
+        const source = baselines.find((b) => b.id === wantedId) ?? baselines.find((b) => b.agentKind === agentKind);
+        if (!source)
+            throw new ValidationError(`No baseline prompt for agent kind "${agentKind}"`);
+        return source;
+    }
+    /** Seed the builtin fixture library for a workspace that has none yet. Idempotent. */
+    async ensureBuiltinFixtures(workspaceId) {
+        const current = await this.deps.sandboxFixtureRepository.list(workspaceId);
+        if (current.length > 0)
+            return;
+        for (const fixture of listBuiltinFixtures(this.deps.clock.now())) {
+            await this.deps.sandboxFixtureRepository.upsert(workspaceId, fixture);
+        }
+    }
+    /**
+     * The judge model to use when the caller didn't pick one: the deployment's routing
+     * default. We do NOT guess a provider — if no default is configured (e.g. a minimal
+     * deployment), require an explicit `judgeModel` at create time rather than defaulting to
+     * a vendor that may have no key, which would otherwise fail every cell's grade at launch.
+     */
+    defaultJudgeModel() {
+        const ref = this.deps.defaultModelRef;
+        if (!ref) {
+            throw new ValidationError('No default model is configured for the Sandbox judge; specify judgeModel explicitly.');
+        }
+        return `${ref.provider}:${ref.model}`;
+    }
+}
+//# sourceMappingURL=SandboxService.js.map

package/dist/modules/sandbox/SandboxService.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"SandboxService.js","sourceRoot":"","sources":["../../../src/modules/sandbox/SandboxService.ts"],"names":[],"mappings":"AAgBA,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAA;AAQnG,OAAO,EACL,SAAS,EACT,wBAAwB,EACxB,gBAAgB,EAChB,aAAa,EACb,mBAAmB,EACnB,WAAW,EACX,mBAAmB,EAEnB,iBAAiB,EACjB,eAAe,GAChB,MAAM,sBAAsB,CAAA;AAE7B,oFAAoF;AACpF,MAAM,CAAC,MAAM,iBAAiB,GAAG,GAAG,CAAA;AAgBpC;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,KAAgC,EAChC,WAAmB,EACnB,YAAoB;IAEpB,MAAM,UAAU,GAAG,WAAW,CAC5B,MAAM,KAAK,CAAC,2BAA2B,CAAC,GAAG,CAAC,WAAW,EAAE,YAAY,CAAC,EACtE,mBAAmB,EACnB,YAAY,CACb,CAAA;IACD,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QACvC,KAAK,CAAC,oBAAoB,CAAC,gBAAgB,CAAC,WAAW,EAAE,YAAY,CAAC;QACtE,KAAK,CAAC,sBAAsB,CAAC,gBAAgB,CAAC,WAAW,EAAE,YAAY,CAAC;KACzE,CAAC,CAAA;IACF,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,EAAE,CAAA;AACrC,CAAC;AA6BD;;;;;;GAMG;AACH,MAAM,OAAO,cAAc;IACI,IAAI;IAAjC,YAA6B,IAAgC;oBAAhC,IAAI;IAA+B,CAAC;IAEjE,4EAA4E;IAC5E,KAAK,CAAC,QAAQ,CAAC,WAAmB;QAChC,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YACzD,IAAI,CAAC,WAAW,CAAC,WAAW,CAAC;YAC7B,IAAI,CAAC,YAAY,CAAC,WAAW,CAAC;YAC9B,IAAI,CAAC,IAAI,CAAC,2BAA2B,CAAC,IAAI,CAAC,WAAW,CAAC;SACxD,CAAC,CAAA;QACF,OAAO;YACL,UAAU,EAAE,mBAAmB;YAC/B,OAAO;YACP,QAAQ;YACR,WAAW;YACX,QAAQ,EAAE,iBAAiB;SAC5B,CAAA;IACH,CAAC;IAED,8EAA8E;IAE9E,+EAA+E;IAC/E,KAAK,CAAC,WAAW,CAAC,WAAmB,EAAE,SAAkB;QACvD,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAA;QACtD,MAAM,UAAU,GAAG,SAAS;YAC1B,CAAC,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,UAAU,CAAC,WAAW,EAAE,SAAS,CAAC;YACnF,CAAC,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,IAAI,CAAC,WAAW,CAAC,CAAA;QACpE,MAAM,SAAS,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAA;QAC5F,OAAO,CAAC,GAAG,SAAS,EAAE,GAAG,UAAU,CAAC,CAAA;IACtC,CAAC;IAED,qFAAqF;IACrF,KAAK,CAAC,WAAW,CACf,WAAmB,EACnB,KAA8B;QAE9B,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,MAAM,GAAG,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,YAAY,CAAC,CAAA;QACxE,MAAM,IAAI,GAAG,eAAe,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;QAC7C,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,IAAI,IAAI,EAAE,KAAK,IAAI,KAAK,CAAC,SAAS,CAAA;QACzD,MAAM,OAAO,GAAG,wBAAwB,CACtC;YACE,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,YAAY,EAAE,MAAM,CAAC,YAAY;SAClC,EACD,IAAI,EACJ;YACE,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC;YACrC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;YAChC,SAAS,EAAE,IAAI;YACf,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,EAAE;SAC3B,CACF,CAAA;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QAC3E,OAAO,OAAO,CAAA;IAChB,CAAC;IAED,sGAAsG;IACtG,KAAK,CAAC,WAAW,CACf,WAAmB,EACnB,KAA8B;QAE9B,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,MAAM,GAAG;YACb,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC;YACrC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;YAChC,SAAS,EAAE,IAAI;YACf,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,EAAE;SAC3B,CAAA;QACD,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;YAC3C,MAAM,QAAQ,GAAG,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,QAAQ,CAAC,CAAA;YAC1F,IAAI,CAAC,QAAQ;gBAAE,MAAM,IAAI,eAAe,CAAC,4BAA4B,KAAK,CAAC,QAAQ,GAAG,CAAC,CAAA;YACvF,MAAM,IAAI,GAAG,eAAe,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAA;YAChD,MAAM,OAAO,GAAG,wBAAwB,CACtC;gBACE,SAAS,EAAE,QAAQ,CAAC,SAAS;gBAC7B,UAAU,EAAE,KAAK,CAAC,UAAU;gBAC5B,YAAY,EAAE,QAAQ,CAAC,YAAY;aACpC,EACD,IAAI,EAAE,KAAK,IAAI,QAAQ,CAAC,SAAS,EACjC,MAAM,CACP,CAAA;YACD,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;YAC3E,OAAO,OAAO,CAAA;QAChB,CAAC;QACD,MAAM,MAAM,GAAG,WAAW,CACxB,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,QAAQ,CAAC,EAC/E,sBAAsB,EACtB,KAAK,CAAC,QAAQ,CACf,CAAA;QACD,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,EAAE,KAAK,CAAC,UAAU,EAAE,MAAM,CAAC,CAAA;QAC7D,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QAC3E,OAAO,OAAO,CAAA;IAChB,CAAC;IAED,4CAA4C;IAC5C,KAAK,CAAC,SAAS,CACb,WAAmB,EACnB,EAAU,EACV,KAA4B;QAE5B,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,QAAQ,GAAG,WAAW,CAC1B,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,GAAG,CAAC,WAAW,EAAE,EAAE,CAAC,EACnE,sBAAsB,EACtB,EAAE,CACH,CAAA;QACD,MAAM,OAAO,GAAG,EAAE,GAAG,QAAQ,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAA;QACrD,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QAC3E,OAAO,OAAO,CAAA;IAChB,CAAC;IAED,0EAA0E;IAC1E,KAAK,CAAC,aAAa,CAAC,WAAmB,EAAE,EAAU;QACjD,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,IAAI,CAAC,IAAI,CAAC,8BAA8B,CAAC,OAAO,CAAC,WAAW,EAAE,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAA;IAChG,CAAC;IAED,8EAA8E;IAE9E,sEAAsE;IACtE,KAAK,CAAC,YAAY,CAAC,WAAmB;QACpC,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,IAAI,CAAC,qBAAqB,CAAC,WAAW,CAAC,CAAA;QAC7C,OAAO,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAA;IAC7D,CAAC;IAED,2CAA2C;IAC3C,KAAK,CAAC,aAAa,CACjB,WAAmB,EACnB,KAAgC;QAEhC,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,OAAO,GAAmB;YAC9B,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC;YACrC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,OAAO,EAAE,KAAK,CAAC,OAAO,IAAI,IAAI;YAC9B,OAAO,EAAE,KAAK,CAAC,OAAO,IAAI,IAAI;YAC9B,SAAS,EAAE,KAAK,CAAC,SAAS,IAAI,IAAI;YAClC,MAAM,EAAE,QAAQ;YAChB,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;SACjC,CAAA;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACrE,OAAO,OAAO,CAAA;IAChB,CAAC;IAED,+EAA+E;IAC/E,KAAK,CAAC,aAAa,CAAC,WAAmB,EAAE,EAAU;QACjD,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,GAAG,CAAC,WAAW,EAAE,EAAE,CAAC,CAAA;QAC9E,IAAI,QAAQ,EAAE,MAAM,KAAK,SAAS,EAAE,CAAC;YACnC,MAAM,IAAI,aAAa,CAAC,qCAAqC,CAAC,CAAA;QAChE,CAAC;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,MAAM,CAAC,WAAW,EAAE,EAAE,CAAC,CAAA;IAClE,CAAC;IAED,8EAA8E;IAE9E,KAAK,CAAC,eAAe,CAAC,WAAmB;QACvC,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,OAAO,IAAI,CAAC,IAAI,CAAC,2BAA2B,CAAC,IAAI,CAAC,WAAW,CAAC,CAAA;IAChE,CAAC;IAED,2DAA2D;IAC3D,KAAK,CAAC,aAAa,CAAC,WAAmB,EAAE,EAAU;QACjD,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,OAAO,uBAAuB,CAAC,IAAI,CAAC,IAAI,EAAE,WAAW,EAAE,EAAE,CAAC,CAAA;IAC5D,CAAC;IAED,4FAA4F;IAC5F,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,KAAmC;QAEnC,MAAM,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAClE,MAAM,IAAI,GAAG,eAAe,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;QAC7C,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,MAAM,IAAI,eAAe,CAAC,IAAI,KAAK,CAAC,SAAS,wCAAwC,CAAC,CAAA;QACxF,CAAC;QACD,qFAAqF;QACrF,oFAAoF;QACpF,oEAAoE;QACpE,IAAI,IAAI,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAChC,MAAM,IAAI,eAAe,CACvB,QAAQ,KAAK,CAAC,SAAS,0FAA0F,CAClH,CAAA;QACH,CAAC;QACD,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;YACpC,MAAM,IAAI,eAAe,CACvB,oEAAoE,CACrE,CAAA;QACH,CAAC;QACD,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,IAAI,CAAC,CAAA;QAClC,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;QAC9C,IAAI,KAAK,GAAG,iBAAiB,EAAE,CAAC;YAC9B,MAAM,IAAI,eAAe,CACvB,0BAA0B,KAAK,wBAAwB,iBAAiB,yBAAyB,CAClG,CAAA;QACH,CAAC;QACD,MAAM,UAAU,GAAsB;YACpC,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC;YACrC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,UAAU,EAAE,KAAK,CAAC,UAAU,IAAI,IAAI,CAAC,iBAAiB,EAAE;YACxD,OAAO;YACP,MAAM,EAAE,OAAO;YACf,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,YAAY,EAAE,KAAK,CAAC,YAAY,IAAI,IAAI;YACxC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;YAChC,SAAS,EAAE,IAAI;SAChB,CAAA;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,2BAA2B,CAAC,MAAM,CAAC,WAAW,EAAE,UAAU,CAAC,CAAA;QAC3E,OAAO,UAAU,CAAA;IACnB,CAAC;IAED,8EAA8E;IAE9E,2FAA2F;IACnF,eAAe,CAAC,SAAiB,EAAE,YAA2B;QACpE,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAA;QACtD,MAAM,QAAQ,GAAG,YAAY,CAAC,CAAC,CAAC,YAAY,YAAY,EAAE,CAAC,CAAC,CAAC,iBAAiB,CAAC,SAAS,CAAC,CAAA;QACzF,MAAM,MAAM,GACV,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,QAAQ,CAAC,IAAI,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAA;QAC9F,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,eAAe,CAAC,sCAAsC,SAAS,GAAG,CAAC,CAAA;QAC1F,OAAO,MAAM,CAAA;IACf,CAAC;IAED,sFAAsF;IAC9E,KAAK,CAAC,qBAAqB,CAAC,WAAmB;QACrD,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAA;QAC1E,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC;YAAE,OAAM;QAC9B,KAAK,MAAM,OAAO,IAAI,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC;YACjE,MAAM,IAAI,CAAC,IAAI,CAAC,wBAAwB,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACvE,CAAC;IACH,CAAC;IAED;;;;;OAKG;IACK,iBAAiB;QACvB,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,eAAe,CAAA;QACrC,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,MAAM,IAAI,eAAe,CACvB,sFAAsF,CACvF,CAAA;QACH,CAAC;QACD,OAAO,GAAG,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,KAAK,EAAE,CAAA;IACvC,CAAC;CACF"}

package/dist/modules/sandbox/sandbox.logic.d.ts

@@ -0,0 +1,37 @@
+import type { SandboxExpectation, SandboxFixture, SandboxGradeDimension, SandboxObjectiveResult } from '@cat-factory/contracts';
+import type { ModelRef } from '@cat-factory/kernel';
+import { type ExpectationScore, type Rubric } from '@cat-factory/sandbox';
+export { extractJson } from '@cat-factory/kernel';
+/** Split a model catalog id (`provider:model`) into a {@link ModelRef}. */
+export declare function parseModelCatalogId(id: string): ModelRef;
+/**
+ * Render an inline fixture's payload into the task input the candidate reasons over (its
+ * system prompt carries the role instructions) and the judge grades against. Defensive:
+ * the payload is a `Record<string, unknown>` matching a `RequirementsContext` /
+ * `ClarityContext` (a `block` + optional `docs`/`tasks`) or a reviewer `AgentRunContext`
+ * (a `block` + the work-to-review in `priorOutputs`), so it reads each field tolerantly.
+ */
+export declare function renderFixtureInput(fixture: SandboxFixture): string;
+/** System prompt for the Sandbox judge — a reference-free rubric grader. */
+export declare const JUDGE_SYSTEM_PROMPT: string;
+/** Build the judge user prompt for one cell: rubric + task input + candidate output + expectations. */
+export declare function buildJudgePrompt(rubric: Rubric, taskInput: string, output: string, expectations: readonly SandboxExpectation[]): string;
+/**
+ * Coerce a judge's raw JSON into one score per rubric dimension. Scores are clamped to
+ * [1,5]; a dimension the judge omitted (or scored non-numerically) defaults to 1 with a
+ * note, so the weighted mean never silently drops a dimension from its denominator.
+ */
+export declare function coerceJudgeScores(rubric: Rubric, raw: unknown): SandboxGradeDimension[];
+/** Project the deterministic objective score into the wire `SandboxObjectiveResult` (findings). */
+export declare function toFindingsObjectiveResult(score: ExpectationScore): SandboxObjectiveResult;
+/** Score a candidate's output against a fixture's `findings` objective, if it declares one. */
+export declare function objectiveFor(fixture: SandboxFixture, output: string): SandboxObjectiveResult | null;
+/**
+ * How many rubric dimensions the judge actually scored with a usable numeric value. The
+ * run-driver treats a count of 0 as a grading FAILURE (record an error on the cell) rather
+ * than letting {@link coerceJudgeScores} silently floor every dimension to 1 — an
+ * unparseable / empty / reasoning-only judge reply must not masquerade as a confident
+ * bottom-of-scale grade.
+ */
+export declare function gradedDimensionCount(rubric: Rubric, raw: unknown): number;
+//# sourceMappingURL=sandbox.logic.d.ts.map

package/dist/modules/sandbox/sandbox.logic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox.logic.d.ts","sourceRoot":"","sources":["../../../src/modules/sandbox/sandbox.logic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,kBAAkB,EAClB,cAAc,EACd,qBAAqB,EACrB,sBAAsB,EACvB,MAAM,wBAAwB,CAAA;AAC/B,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAA;AAEnD,OAAO,EACL,KAAK,gBAAgB,EACrB,KAAK,MAAM,EAGZ,MAAM,sBAAsB,CAAA;AAK7B,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAA;AAOjD,2EAA2E;AAC3E,wBAAgB,mBAAmB,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,CAIxD;AAED;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,cAAc,GAAG,MAAM,CAyClE;AAED,4EAA4E;AAC5E,eAAO,MAAM,mBAAmB,QAUrB,CAAA;AAEX,uGAAuG;AACvG,wBAAgB,gBAAgB,CAC9B,MAAM,EAAE,MAAM,EACd,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,MAAM,EACd,YAAY,EAAE,SAAS,kBAAkB,EAAE,GAC1C,MAAM,CAsBR;AAED;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,GAAG,qBAAqB,EAAE,CAavF;AAED,mGAAmG;AACnG,wBAAgB,yBAAyB,CAAC,KAAK,EAAE,gBAAgB,GAAG,sBAAsB,CAWzF;AAED,+FAA+F;AAC/F,wBAAgB,YAAY,CAC1B,OAAO,EAAE,cAAc,EACvB,MAAM,EAAE,MAAM,GACb,sBAAsB,GAAG,IAAI,CAI/B;AAED;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,GAAG,MAAM,CAOzE"}

package/dist/modules/sandbox/sandbox.logic.js

@@ -0,0 +1,178 @@
+import { FINAL_ANSWER_IN_REPLY } from '@cat-factory/agents';
+import { renderExpectationBrief, scoreExpectations, } from '@cat-factory/sandbox';
+// The robust LLM-reply JSON extractor lives in the kernel (one copy shared by the
+// requirements reviewer, the document planner, and the Sandbox judge). Re-exported here
+// so the run-driver imports it alongside the other judge helpers.
+export { extractJson } from '@cat-factory/kernel';
+// Pure helpers for the Sandbox run-driver + judge. Kept side-effect-free (no LLM/IO,
+// no clock/identity) so the candidate-input rendering, judge-prompt assembly, score
+// coercion and objective projection are deterministic and unit-testable; the service
+// wraps them with the model-provider calls + persistence.
+/** Split a model catalog id (`provider:model`) into a {@link ModelRef}. */
+export function parseModelCatalogId(id) {
+    const idx = id.indexOf(':');
+    if (idx === -1)
+        return { provider: id, model: '' };
+    return { provider: id.slice(0, idx), model: id.slice(idx + 1) };
+}
+/**
+ * Render an inline fixture's payload into the task input the candidate reasons over (its
+ * system prompt carries the role instructions) and the judge grades against. Defensive:
+ * the payload is a `Record<string, unknown>` matching a `RequirementsContext` /
+ * `ClarityContext` (a `block` + optional `docs`/`tasks`) or a reviewer `AgentRunContext`
+ * (a `block` + the work-to-review in `priorOutputs`), so it reads each field tolerantly.
+ */
+export function renderFixtureInput(fixture) {
+    const payload = (fixture.payload ?? {});
+    const parts = [];
+    const block = payload.block;
+    if (block) {
+        const heading = block.type ? `${block.title ?? 'Untitled'} (${block.type})` : block.title;
+        parts.push(`# ${heading ?? 'Untitled'}`);
+        if (block.description)
+            parts.push(block.description);
+    }
+    const docs = Array.isArray(payload.docs) ? payload.docs : [];
+    if (docs.length > 0) {
+        parts.push('## Linked documents');
+        for (const doc of docs) {
+            const d = doc;
+            parts.push(`### ${d.title ?? 'Document'}\n${d.body ?? d.content ?? ''}`.trim());
+        }
+    }
+    const tasks = Array.isArray(payload.tasks) ? payload.tasks : [];
+    if (tasks.length > 0) {
+        parts.push('## Linked tracker issues');
+        for (const task of tasks) {
+            const t = task;
+            parts.push(`- ${t.title ?? 'Issue'}${t.body ? `: ${t.body}` : ''}`);
+        }
+    }
+    const priorOutputs = Array.isArray(payload.priorOutputs)
+        ? payload.priorOutputs
+        : [];
+    if (priorOutputs.length > 0) {
+        parts.push('## Work from earlier agents');
+        for (const prior of priorOutputs) {
+            const p = prior;
+            parts.push(`### ${p.agentKind ?? 'agent'}\n${p.output ?? ''}`.trim());
+        }
+    }
+    return parts.join('\n\n').trim();
+}
+/** System prompt for the Sandbox judge — a reference-free rubric grader. */
+export const JUDGE_SYSTEM_PROMPT = [
+    "You are a meticulous, impartial evaluator. You grade an AI agent's output for a given",
+    'task against a fixed rubric, scoring each dimension from 1 (poor) to 5 (excellent).',
+    'Judge ONLY against the task input and the candidate output you are given — never invent',
+    'context. Be calibrated: reserve 5 for genuinely excellent work and 1 for output that',
+    'fails the dimension outright. Your entire visible reply MUST be the requested JSON object',
+    'and nothing else.',
+    // The judge's deliverable IS its (parsed) final reply, so append the shared directive
+    // that keeps reasoning models from emitting the answer into a hidden thinking channel.
+    FINAL_ANSWER_IN_REPLY,
+].join(' ');
+/** Build the judge user prompt for one cell: rubric + task input + candidate output + expectations. */
+export function buildJudgePrompt(rubric, taskInput, output, expectations) {
+    const dims = rubric.dimensions
+        .map((d) => `- "${d.key}" — ${d.label}: ${d.description}`)
+        .join('\n');
+    const brief = renderExpectationBrief(expectations);
+    return [
+        `You are grading an AI agent's output for a "${rubric.task}" task.`,
+        '',
+        '## Task input',
+        taskInput || '(no task input was supplied)',
+        '',
+        '## Candidate output',
+        output.trim() || '(the candidate produced no output)',
+        '',
+        '## Rubric — score every dimension from 1 to 5',
+        dims,
+        ...(brief ? ['', brief] : []),
+        '',
+        'Respond with ONLY this JSON object (no prose, no code fences):',
+        '{"scores":[{"key":"<dimension key>","score":<1-5>,"rationale":"<one short sentence>"}]}',
+        'Include every dimension key exactly once.',
+    ].join('\n');
+}
+/**
+ * Coerce a judge's raw JSON into one score per rubric dimension. Scores are clamped to
+ * [1,5]; a dimension the judge omitted (or scored non-numerically) defaults to 1 with a
+ * note, so the weighted mean never silently drops a dimension from its denominator.
+ */
+export function coerceJudgeScores(rubric, raw) {
+    const byKey = scoreEntriesByKey(raw);
+    return rubric.dimensions.map((dim) => {
+        const found = byKey.get(dim.key);
+        const score = clampScore(found?.score);
+        const rationale = typeof found?.rationale === 'string' && found.rationale.trim()
+            ? found.rationale.trim()
+            : score === null
+                ? 'Judge did not score this dimension.'
+                : '';
+        return { key: dim.key, score: score ?? 1, rationale };
+    });
+}
+/** Project the deterministic objective score into the wire `SandboxObjectiveResult` (findings). */
+export function toFindingsObjectiveResult(score) {
+    return {
+        kind: 'findings',
+        pass: score.missedHighImpact.length === 0,
+        detail: `Caught ${score.caught.length}/${score.caught.length + score.missed.length} expected findings; impact recall ${score.impactRecall}, wow bonus ${score.wowBonus}.`,
+        impactRecall: score.impactRecall,
+        wowBonus: score.wowBonus,
+        caught: score.caught.length,
+        total: score.caught.length + score.missed.length,
+        missedHighImpact: score.missedHighImpact,
+    };
+}
+/** Score a candidate's output against a fixture's `findings` objective, if it declares one. */
+export function objectiveFor(fixture, output) {
+    const objective = fixture.objective;
+    if (!objective || objective.kind !== 'findings')
+        return null;
+    return toFindingsObjectiveResult(scoreExpectations(objective.expectations, output));
+}
+/**
+ * How many rubric dimensions the judge actually scored with a usable numeric value. The
+ * run-driver treats a count of 0 as a grading FAILURE (record an error on the cell) rather
+ * than letting {@link coerceJudgeScores} silently floor every dimension to 1 — an
+ * unparseable / empty / reasoning-only judge reply must not masquerade as a confident
+ * bottom-of-scale grade.
+ */
+export function gradedDimensionCount(rubric, raw) {
+    const byKey = scoreEntriesByKey(raw);
+    let count = 0;
+    for (const dim of rubric.dimensions) {
+        if (clampScore(byKey.get(dim.key)?.score) !== null)
+            count++;
+    }
+    return count;
+}
+function scoreEntriesByKey(raw) {
+    const byKey = new Map();
+    for (const entry of extractScoreArray(raw)) {
+        if (entry &&
+            typeof entry === 'object' &&
+            typeof entry.key === 'string') {
+            byKey.set(entry.key, entry);
+        }
+    }
+    return byKey;
+}
+function extractScoreArray(raw) {
+    if (Array.isArray(raw))
+        return raw;
+    if (raw && typeof raw === 'object' && Array.isArray(raw.scores)) {
+        return raw.scores;
+    }
+    return [];
+}
+function clampScore(value) {
+    const n = typeof value === 'number' ? value : typeof value === 'string' ? Number(value) : NaN;
+    if (!Number.isFinite(n))
+        return null;
+    return Math.min(5, Math.max(1, n));
+}
+//# sourceMappingURL=sandbox.logic.js.map

package/dist/modules/sandbox/sandbox.logic.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox.logic.js","sourceRoot":"","sources":["../../../src/modules/sandbox/sandbox.logic.ts"],"names":[],"mappings":"AAOA,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAA;AAC3D,OAAO,EAGL,sBAAsB,EACtB,iBAAiB,GAClB,MAAM,sBAAsB,CAAA;AAE7B,kFAAkF;AAClF,wFAAwF;AACxF,kEAAkE;AAClE,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAA;AAEjD,qFAAqF;AACrF,oFAAoF;AACpF,qFAAqF;AACrF,0DAA0D;AAE1D,2EAA2E;AAC3E,MAAM,UAAU,mBAAmB,CAAC,EAAU;IAC5C,MAAM,GAAG,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAC3B,IAAI,GAAG,KAAK,CAAC,CAAC;QAAE,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,CAAA;IAClD,OAAO,EAAE,QAAQ,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,EAAE,CAAA;AACjE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,kBAAkB,CAAC,OAAuB;IACxD,MAAM,OAAO,GAAG,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,CAA4B,CAAA;IAClE,MAAM,KAAK,GAAa,EAAE,CAAA;IAE1B,MAAM,KAAK,GAAG,OAAO,CAAC,KAA4E,CAAA;IAClG,IAAI,KAAK,EAAE,CAAC;QACV,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,IAAI,UAAU,KAAK,KAAK,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAA;QACzF,KAAK,CAAC,IAAI,CAAC,KAAK,OAAO,IAAI,UAAU,EAAE,CAAC,CAAA;QACxC,IAAI,KAAK,CAAC,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,CAAA;IACtD,CAAC;IAED,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAE,OAAO,CAAC,IAAkB,CAAC,CAAC,CAAC,EAAE,CAAA;IAC3E,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpB,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAA;QACjC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,CAAC,GAAG,GAA0D,CAAA;YACpE,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,IAAI,UAAU,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,CAAA;QACjF,CAAC;IACH,CAAC;IAED,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAE,OAAO,CAAC,KAAmB,CAAC,CAAC,CAAC,EAAE,CAAA;IAC9E,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrB,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAA;QACtC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,CAAC,GAAG,IAAyC,CAAA;YACnD,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,IAAI,OAAO,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAA;QACrE,CAAC;IACH,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,YAAY,CAAC;QACtD,CAAC,CAAE,OAAO,CAAC,YAA0B;QACrC,CAAC,CAAC,EAAE,CAAA;IACN,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5B,KAAK,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAA;QACzC,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;YACjC,MAAM,CAAC,GAAG,KAAgD,CAAA;YAC1D,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,SAAS,IAAI,OAAO,KAAK,CAAC,CAAC,MAAM,IAAI,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,CAAA;QACvE,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAA;AAClC,CAAC;AAED,4EAA4E;AAC5E,MAAM,CAAC,MAAM,mBAAmB,GAAG;IACjC,uFAAuF;IACvF,qFAAqF;IACrF,yFAAyF;IACzF,sFAAsF;IACtF,2FAA2F;IAC3F,mBAAmB;IACnB,sFAAsF;IACtF,uFAAuF;IACvF,qBAAqB;CACtB,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;AAEX,uGAAuG;AACvG,MAAM,UAAU,gBAAgB,CAC9B,MAAc,EACd,SAAiB,EACjB,MAAc,EACd,YAA2C;IAE3C,MAAM,IAAI,GAAG,MAAM,CAAC,UAAU;SAC3B,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC;SACzD,IAAI,CAAC,IAAI,CAAC,CAAA;IACb,MAAM,KAAK,GAAG,sBAAsB,CAAC,YAAY,CAAC,CAAA;IAClD,OAAO;QACL,+CAA+C,MAAM,CAAC,IAAI,SAAS;QACnE,EAAE;QACF,eAAe;QACf,SAAS,IAAI,8BAA8B;QAC3C,EAAE;QACF,qBAAqB;QACrB,MAAM,CAAC,IAAI,EAAE,IAAI,oCAAoC;QACrD,EAAE;QACF,+CAA+C;QAC/C,IAAI;QACJ,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7B,EAAE;QACF,gEAAgE;QAChE,yFAAyF;QACzF,2CAA2C;KAC5C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACd,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAc,EAAE,GAAY;IAC5D,MAAM,KAAK,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAA;IACpC,OAAO,MAAM,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE;QACnC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QAChC,MAAM,KAAK,GAAG,UAAU,CAAC,KAAK,EAAE,KAAK,CAAC,CAAA;QACtC,MAAM,SAAS,GACb,OAAO,KAAK,EAAE,SAAS,KAAK,QAAQ,IAAI,KAAK,CAAC,SAAS,CAAC,IAAI,EAAE;YAC5D,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,EAAE;YACxB,CAAC,CAAC,KAAK,KAAK,IAAI;gBACd,CAAC,CAAC,qCAAqC;gBACvC,CAAC,CAAC,EAAE,CAAA;QACV,OAAO,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK,IAAI,CAAC,EAAE,SAAS,EAAE,CAAA;IACvD,CAAC,CAAC,CAAA;AACJ,CAAC;AAED,mGAAmG;AACnG,MAAM,UAAU,yBAAyB,CAAC,KAAuB;IAC/D,OAAO;QACL,IAAI,EAAE,UAAU;QAChB,IAAI,EAAE,KAAK,CAAC,gBAAgB,CAAC,MAAM,KAAK,CAAC;QACzC,MAAM,EAAE,UAAU,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,qCAAqC,KAAK,CAAC,YAAY,eAAe,KAAK,CAAC,QAAQ,GAAG;QACzK,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,MAAM;QAC3B,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM;QAChD,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;KACzC,CAAA;AACH,CAAC;AAED,+FAA+F;AAC/F,MAAM,UAAU,YAAY,CAC1B,OAAuB,EACvB,MAAc;IAEd,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAA;IACnC,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,IAAI,KAAK,UAAU;QAAE,OAAO,IAAI,CAAA;IAC5D,OAAO,yBAAyB,CAAC,iBAAiB,CAAC,SAAS,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC,CAAA;AACrF,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,oBAAoB,CAAC,MAAc,EAAE,GAAY;IAC/D,MAAM,KAAK,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAA;IACpC,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;QACpC,IAAI,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,KAAK,IAAI;YAAE,KAAK,EAAE,CAAA;IAC7D,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,SAAS,iBAAiB,CAAC,GAAY;IACrC,MAAM,KAAK,GAAG,IAAI,GAAG,EAAoD,CAAA;IACzE,KAAK,MAAM,KAAK,IAAI,iBAAiB,CAAC,GAAG,CAAC,EAAE,CAAC;QAC3C,IACE,KAAK;YACL,OAAO,KAAK,KAAK,QAAQ;YACzB,OAAQ,KAA2B,CAAC,GAAG,KAAK,QAAQ,EACpD,CAAC;YACD,KAAK,CAAC,GAAG,CAAE,KAAyB,CAAC,GAAG,EAAE,KAAiD,CAAC,CAAA;QAC9F,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,SAAS,iBAAiB,CAAC,GAAY;IACrC,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC;QAAE,OAAO,GAAG,CAAA;IAClC,IAAI,GAAG,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAE,GAA4B,CAAC,MAAM,CAAC,EAAE,CAAC;QAC1F,OAAQ,GAA6B,CAAC,MAAM,CAAA;IAC9C,CAAC;IACD,OAAO,EAAE,CAAA;AACX,CAAC;AAED,SAAS,UAAU,CAAC,KAAc;IAChC,MAAM,CAAC,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAA;IAC7F,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;QAAE,OAAO,IAAI,CAAA;IACpC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;AACpC,CAAC"}

package/dist/modules/settings/WorkspaceSettingsService.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"WorkspaceSettingsService.d.ts","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,4BAA4B,EAC5B,mBAAmB,EACnB,iBAAiB,EACjB,2BAA2B,EAC5B,MAAM,qBAAqB,CAAA;AAG5B,MAAM,WAAW,oCAAoC;IACnD,2BAA2B,EAAE,2BAA2B,CAAA;IACxD,mBAAmB,EAAE,mBAAmB,CAAA;CACzC;AAED;;;;;;GAMG;AACH,qBAAa,wBAAwB;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IAEzD,YAAY,IAAI,EAAE,oCAAoC,EAGrD;IAED,0FAA0F;IACpF,GAAG,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAEzD;IAED,kEAAkE;IAC5D,MAAM,CACV,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,4BAA4B,GAClC,OAAO,CAAC,iBAAiB,CAAC,CAyB5B;CACF"}
+{"version":3,"file":"WorkspaceSettingsService.d.ts","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,4BAA4B,EAC5B,mBAAmB,EACnB,iBAAiB,EACjB,2BAA2B,EAC5B,MAAM,qBAAqB,CAAA;AAG5B,MAAM,WAAW,oCAAoC;IACnD,2BAA2B,EAAE,2BAA2B,CAAA;IACxD,mBAAmB,EAAE,mBAAmB,CAAA;CACzC;AAED;;;;;;GAMG;AACH,qBAAa,wBAAwB;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IAEzD,YAAY,IAAI,EAAE,oCAAoC,EAGrD;IAED,0FAA0F;IACpF,GAAG,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAEzD;IAED,kEAAkE;IAC5D,MAAM,CACV,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,4BAA4B,GAClC,OAAO,CAAC,iBAAiB,CAAC,CA+B5B;CACF"}

package/dist/modules/settings/WorkspaceSettingsService.js

@@ -26,6 +26,9 @@ export class WorkspaceSettingsService {
             taskLimitMode: patch.taskLimitMode ?? current.taskLimitMode,
             taskLimitShared: patch.taskLimitShared !== undefined ? patch.taskLimitShared : current.taskLimitShared,
             taskLimitPerType: patch.taskLimitPerType !== undefined ? patch.taskLimitPerType : current.taskLimitPerType,
+            spendCurrency: patch.spendCurrency !== undefined ? patch.spendCurrency : current.spendCurrency,
+            spendMonthlyLimit: patch.spendMonthlyLimit !== undefined ? patch.spendMonthlyLimit : current.spendMonthlyLimit,
+            spendModelPrices: patch.spendModelPrices !== undefined ? patch.spendModelPrices : current.spendModelPrices,
         };
         // Keep the limit fields consistent with the mode so the enforcement logic + UI never
         // read a stale cap from an inactive mode.

package/dist/modules/settings/WorkspaceSettingsService.js.map

@@ -1 +1 @@
-{"version":3,"file":"WorkspaceSettingsService.js","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,0BAA0B,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAA;AAOlF;;;;;;GAMG;AACH,MAAM,OAAO,wBAAwB;IAClB,QAAQ,CAA6B;IACrC,mBAAmB,CAAqB;IAEzD,YAAY,IAA0C;QACpD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,2BAA2B,CAAA;QAChD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,CAAA;IACrD,CAAC;IAED,0FAA0F;IAC1F,KAAK,CAAC,GAAG,CAAC,WAAmB;QAC3B,OAAO,CAAC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,GAAG,0BAA0B,EAAE,CAAA;IACpF,CAAC;IAED,kEAAkE;IAClE,KAAK,CAAC,MAAM,CACV,WAAmB,EACnB,KAAmC;QAEnC,MAAM,gBAAgB,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAC7D,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC3C,MAAM,IAAI,GAAsB;YAC9B,wBAAwB,EAAE,KAAK,CAAC,wBAAwB,IAAI,OAAO,CAAC,wBAAwB;YAC5F,aAAa,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,aAAa;YAC3D,eAAe,EACb,KAAK,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe;YACvF,gBAAgB,EACd,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB;SAC3F,CAAA;QACD,qFAAqF;QACrF,0CAA0C;QAC1C,IAAI,IAAI,CAAC,aAAa,KAAK,KAAK,EAAE,CAAC;YACjC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;QAC9B,CAAC;aAAM,IAAI,IAAI,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;YAC3C,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;YAC5B,IAAI,IAAI,CAAC,eAAe,IAAI,IAAI;gBAAE,IAAI,CAAC,eAAe,GAAG,CAAC,CAAA;QAC5D,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAAE,IAAI,CAAC,gBAAgB,GAAG,EAAE,CAAA;QAC/D,CAAC;QACD,MAAM,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,IAAI,CAAC,CAAA;QAC7C,OAAO,IAAI,CAAA;IACb,CAAC;CACF"}
+{"version":3,"file":"WorkspaceSettingsService.js","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,0BAA0B,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAA;AAOlF;;;;;;GAMG;AACH,MAAM,OAAO,wBAAwB;IAClB,QAAQ,CAA6B;IACrC,mBAAmB,CAAqB;IAEzD,YAAY,IAA0C;QACpD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,2BAA2B,CAAA;QAChD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,CAAA;IACrD,CAAC;IAED,0FAA0F;IAC1F,KAAK,CAAC,GAAG,CAAC,WAAmB;QAC3B,OAAO,CAAC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,GAAG,0BAA0B,EAAE,CAAA;IACpF,CAAC;IAED,kEAAkE;IAClE,KAAK,CAAC,MAAM,CACV,WAAmB,EACnB,KAAmC;QAEnC,MAAM,gBAAgB,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAC7D,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC3C,MAAM,IAAI,GAAsB;YAC9B,wBAAwB,EAAE,KAAK,CAAC,wBAAwB,IAAI,OAAO,CAAC,wBAAwB;YAC5F,aAAa,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,aAAa;YAC3D,eAAe,EACb,KAAK,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe;YACvF,gBAAgB,EACd,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB;YAC1F,aAAa,EACX,KAAK,CAAC,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa;YACjF,iBAAiB,EACf,KAAK,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,OAAO,CAAC,iBAAiB;YAC7F,gBAAgB,EACd,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB;SAC3F,CAAA;QACD,qFAAqF;QACrF,0CAA0C;QAC1C,IAAI,IAAI,CAAC,aAAa,KAAK,KAAK,EAAE,CAAC;YACjC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;QAC9B,CAAC;aAAM,IAAI,IAAI,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;YAC3C,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;YAC5B,IAAI,IAAI,CAAC,eAAe,IAAI,IAAI;gBAAE,IAAI,CAAC,eAAe,GAAG,CAAC,CAAA;QAC5D,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAAE,IAAI,CAAC,gBAAgB,GAAG,EAAE,CAAA;QAC/D,CAAC;QACD,MAAM,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,IAAI,CAAC,CAAA;QAC7C,OAAO,IAAI,CAAA;IACb,CAAC;CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cat-factory/orchestration",
-  "version": "0.19.2",
+  "version": "0.21.0",
   "description": "Delivery-workflow engine for the Agent Architecture Board (execution, bootstrap, pipelines, board, boardScan, requirements, and composition root).",
   "repository": {
     "type": "git",
@@ -25,13 +25,14 @@
   },
   "dependencies": {
     "ai": "^6.0.209",
-    "@cat-factory/agents": "0.14.5",
-    "@cat-factory/contracts": "0.23.0",
-    "@cat-factory/integrations": "0.18.3",
-    "@cat-factory/kernel": "0.26.1",
-    "@cat-factory/prompt-fragments": "0.7.20",
-    "@cat-factory/spend": "0.8.26",
-    "@cat-factory/workspaces": "0.7.32"
+    "@cat-factory/agents": "0.14.7",
+    "@cat-factory/contracts": "0.25.0",
+    "@cat-factory/integrations": "0.20.0",
+    "@cat-factory/kernel": "0.28.0",
+    "@cat-factory/prompt-fragments": "0.7.22",
+    "@cat-factory/sandbox": "0.8.0",
+    "@cat-factory/spend": "0.9.1",
+    "@cat-factory/workspaces": "0.7.34"
   },
   "devDependencies": {
     "typescript": "7.0.1-rc",