@cat-factory/orchestration 0.20.0 → 0.21.0

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/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cat-factory/orchestration",
-  "version": "0.20.0",
+  "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": "0.14.6",
-    "@cat-factory/contracts": "0.24.0",
-    "@cat-factory/integrations": "0.19.0",
-    "@cat-factory/kernel": "0.27.0",
-    "@cat-factory/prompt-fragments": "0.7.21",
-    "@cat-factory/spend": "0.9.0",
-    "@cat-factory/workspaces": "0.7.33"
+    "@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",