@kernel.chat/agent-os - npm Package Compare versions

Comparing version 0.1.0-alpha.0 to 0.2.0-alpha.0

dist/outcomes.d.ts

+/**
+ * Outcomes — rubric-graded self-evaluation with auto-iteration.
+ *
+ * Adopted from Claude Managed Agents' `define_outcome` pattern. A
+ * separate evaluator (running in its own context, typically with a
+ * different model or at least a different prompt) judges the agent's
+ * work product against per-criterion rubric. If unsatisfied, the
+ * agent revises; the loop runs until satisfied or max_iterations.
+ *
+ * agent-os generalizes CMA's pattern in three ways:
+ *  - the evaluator is pluggable (LLM-based, rules-based, hybrid)
+ *  - the rubric supports weighted criteria
+ *  - each iteration's evaluation is content-addressable so the loop
+ *    is replayable in an audit-grade environment
+ */
+export interface RubricCriterion {
+    /** Stable identifier — used in audit logs. */
+    readonly name: string;
+    /** Plain-language description the evaluator scores against. */
+    readonly description: string;
+    /** Optional weight. Defaults to 1.0. */
+    readonly weight?: number;
+}
+export interface Rubric {
+    readonly criteria: ReadonlyArray<RubricCriterion>;
+    /** Weighted-average score threshold for `satisfied` verdict (0.0–1.0).
+     *  Defaults to 0.85. */
+    readonly satisfied_at?: number;
+}
+export interface CriterionScore {
+    readonly name: string;
+    /** 0.0–1.0. Higher is better. */
+    readonly score: number;
+    /** Plain-language feedback explaining the score; used by the agent
+     *  to revise on the next iteration. */
+    readonly feedback: string;
+}
+export interface RubricEvaluation {
+    readonly per_criterion: ReadonlyArray<CriterionScore>;
+    /** Weighted average across criteria. */
+    readonly overall_score: number;
+    readonly verdict: 'satisfied' | 'needs_revision' | 'failed';
+    /** Optional structured feedback the agent should consume on the next
+     *  iteration. The evaluator may synthesize this from per-criterion
+     *  feedback. */
+    readonly revision_guidance?: string;
+}
+export interface RubricEvaluator {
+    evaluate(work_product: string, rubric: Rubric, iteration: number): Promise<RubricEvaluation>;
+}
+export type OutcomeResultCode = 'satisfied' | 'needs_revision' | 'max_iterations_reached' | 'failed' | 'interrupted';
+export interface OutcomeIteration<T> {
+    readonly iteration: number;
+    readonly work: T;
+    readonly work_serialized: string;
+    readonly evaluation: RubricEvaluation;
+    readonly at: string;
+}
+export interface OutcomeResult<T> {
+    readonly code: OutcomeResultCode;
+    readonly iterations_used: number;
+    readonly final_work: T | null;
+    readonly final_evaluation: RubricEvaluation | null;
+    readonly history: ReadonlyArray<OutcomeIteration<T>>;
+}
+export interface WithOutcomeOptions<T> {
+    readonly max_iterations: number;
+    /** Serialize the agent's work product to a string the evaluator can
+     *  judge. Defaults to JSON.stringify. */
+    readonly serialize?: (work: T) => string;
+    /** Abort signal — interrupting mid-iteration returns the
+     *  `interrupted` result code with the history so far. */
+    readonly signal?: AbortSignal;
+}
+/**
+ * Run an attempt → evaluate → revise loop against a rubric.
+ *
+ * `attempt(feedback)` is called once per iteration. On the first call,
+ * `feedback` is `null`; on subsequent calls it's the evaluator's
+ * revision_guidance (or a synthesized summary of per-criterion feedback
+ * if no guidance was provided).
+ *
+ * The loop terminates on:
+ *   - `satisfied`           verdict from the evaluator
+ *   - `failed`              verdict from the evaluator (non-recoverable)
+ *   - max_iterations reached
+ *   - signal.aborted        (interrupted)
+ *   - attempt() or evaluate() throwing — caught and reported as `failed`
+ */
+export declare function withOutcome<T>(rubric: Rubric, attempt: (feedback: string | null, iteration: number) => Promise<T>, evaluator: RubricEvaluator, options: WithOutcomeOptions<T>): Promise<OutcomeResult<T>>;
+/** Helper: weighted-average score. Useful for evaluator implementations. */
+export declare function weightedAverage(scores: ReadonlyArray<CriterionScore>, rubric: Rubric): number;
+/** Helper: classify a score into a verdict against the rubric's threshold. */
+export declare function classify(score: number, rubric: Rubric): 'satisfied' | 'needs_revision';
+/** A trivial evaluator useful for tests — accepts the work product as
+ *  satisfying if it matches a predicate. Production code uses an LLM
+ *  evaluator running in a separate context. */
+export declare function predicateEvaluator(predicate: (work: string) => boolean, rubric: Rubric, feedback_when_failing: string): RubricEvaluator;
+//# sourceMappingURL=outcomes.d.ts.map

dist/outcomes.d.ts.map

+{"version":3,"file":"outcomes.d.ts","sourceRoot":"","sources":["../src/outcomes.ts"],"names":[],"mappings":"AAGA;;;;;;;;;;;;;;GAcG;AAEH,MAAM,WAAW,eAAe;IAC9B,8CAA8C;IAC9C,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,+DAA+D;IAC/D,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,wCAAwC;IACxC,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CACzB;AAED,MAAM,WAAW,MAAM;IACrB,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAC,eAAe,CAAC,CAAA;IACjD;4BACwB;IACxB,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAA;CAC/B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;IACrB,iCAAiC;IACjC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;IACtB;2CACuC;IACvC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;CAC1B;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,aAAa,EAAE,aAAa,CAAC,cAAc,CAAC,CAAA;IACrD,wCAAwC;IACxC,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAA;IAC9B,QAAQ,CAAC,OAAO,EAAE,WAAW,GAAG,gBAAgB,GAAG,QAAQ,CAAA;IAC3D;;oBAEgB;IAChB,QAAQ,CAAC,iBAAiB,CAAC,EAAE,MAAM,CAAA;CACpC;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,CACN,YAAY,EAAE,MAAM,EACpB,MAAM,EAAE,MAAM,EACd,SAAS,EAAE,MAAM,GAChB,OAAO,CAAC,gBAAgB,CAAC,CAAA;CAC7B;AAED,MAAM,MAAM,iBAAiB,GACzB,WAAW,GACX,gBAAgB,GAChB,wBAAwB,GACxB,QAAQ,GACR,aAAa,CAAA;AAEjB,MAAM,WAAW,gBAAgB,CAAC,CAAC;IACjC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAA;IAChB,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAA;IAChC,QAAQ,CAAC,UAAU,EAAE,gBAAgB,CAAA;IACrC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,aAAa,CAAC,CAAC;IAC9B,QAAQ,CAAC,IAAI,EAAE,iBAAiB,CAAA;IAChC,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAA;IAChC,QAAQ,CAAC,UAAU,EAAE,CAAC,GAAG,IAAI,CAAA;IAC7B,QAAQ,CAAC,gBAAgB,EAAE,gBAAgB,GAAG,IAAI,CAAA;IAClD,QAAQ,CAAC,OAAO,EAAE,aAAa,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAA;CACrD;AAED,MAAM,WAAW,kBAAkB,CAAC,CAAC;IACnC,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAA;IAC/B;6CACyC;IACzC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,MAAM,CAAA;IACxC;6DACyD;IACzD,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAA;CAC9B;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAsB,WAAW,CAAC,CAAC,EACjC,MAAM,EAAE,MAAM,EACd,OAAO,EAAE,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,EAAE,SAAS,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EACnE,SAAS,EAAE,eAAe,EAC1B,OAAO,EAAE,kBAAkB,CAAC,CAAC,CAAC,GAC7B,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAuF3B;AAYD,4EAA4E;AAC5E,wBAAgB,eAAe,CAC7B,MAAM,EAAE,aAAa,CAAC,cAAc,CAAC,EACrC,MAAM,EAAE,MAAM,GACb,MAAM,CAWR;AAED,8EAA8E;AAC9E,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,WAAW,GAAG,gBAAgB,CAGtF;AAED;;+CAE+C;AAC/C,wBAAgB,kBAAkB,CAChC,SAAS,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,EACpC,MAAM,EAAE,MAAM,EACd,qBAAqB,EAAE,MAAM,GAC5B,eAAe,CAkBjB"}

dist/outcomes.js

+/**
+ * Run an attempt → evaluate → revise loop against a rubric.
+ *
+ * `attempt(feedback)` is called once per iteration. On the first call,
+ * `feedback` is `null`; on subsequent calls it's the evaluator's
+ * revision_guidance (or a synthesized summary of per-criterion feedback
+ * if no guidance was provided).
+ *
+ * The loop terminates on:
+ *   - `satisfied`           verdict from the evaluator
+ *   - `failed`              verdict from the evaluator (non-recoverable)
+ *   - max_iterations reached
+ *   - signal.aborted        (interrupted)
+ *   - attempt() or evaluate() throwing — caught and reported as `failed`
+ */
+export async function withOutcome(rubric, attempt, evaluator, options) {
+    const serialize = options.serialize ?? ((t) => JSON.stringify(t));
+    const max = Math.max(1, Math.floor(options.max_iterations));
+    const history = [];
+    let feedback = null;
+    let lastWork = null;
+    let lastEval = null;
+    for (let i = 1; i <= max; i++) {
+        if (options.signal?.aborted) {
+            return {
+                code: 'interrupted',
+                iterations_used: i - 1,
+                final_work: lastWork,
+                final_evaluation: lastEval,
+                history,
+            };
+        }
+        let work;
+        try {
+            work = await attempt(feedback, i);
+        }
+        catch (e) {
+            return {
+                code: 'failed',
+                iterations_used: i - 1,
+                final_work: lastWork,
+                final_evaluation: lastEval,
+                history,
+            };
+        }
+        const serialized = serialize(work);
+        let evaluation;
+        try {
+            evaluation = await evaluator.evaluate(serialized, rubric, i);
+        }
+        catch (e) {
+            return {
+                code: 'failed',
+                iterations_used: i,
+                final_work: work,
+                final_evaluation: lastEval,
+                history,
+            };
+        }
+        const iter = {
+            iteration: i,
+            work,
+            work_serialized: serialized,
+            evaluation,
+            at: new Date().toISOString(),
+        };
+        history.push(iter);
+        lastWork = work;
+        lastEval = evaluation;
+        if (evaluation.verdict === 'satisfied') {
+            return {
+                code: 'satisfied',
+                iterations_used: i,
+                final_work: work,
+                final_evaluation: evaluation,
+                history,
+            };
+        }
+        if (evaluation.verdict === 'failed') {
+            return {
+                code: 'failed',
+                iterations_used: i,
+                final_work: work,
+                final_evaluation: evaluation,
+                history,
+            };
+        }
+        feedback =
+            evaluation.revision_guidance ?? synthesizeFeedback(evaluation.per_criterion);
+    }
+    return {
+        code: 'max_iterations_reached',
+        iterations_used: max,
+        final_work: lastWork,
+        final_evaluation: lastEval,
+        history,
+    };
+}
+/** Default feedback synthesizer — joins per-criterion feedback into a
+ *  single message the agent can act on. */
+function synthesizeFeedback(scores) {
+    const weakest = [...scores].sort((a, b) => a.score - b.score).slice(0, 3);
+    if (weakest.length === 0)
+        return '';
+    return weakest
+        .map((s) => `[${s.name} score=${s.score.toFixed(2)}] ${s.feedback}`)
+        .join('\n');
+}
+/** Helper: weighted-average score. Useful for evaluator implementations. */
+export function weightedAverage(scores, rubric) {
+    if (scores.length === 0)
+        return 0;
+    const lookup = new Map(rubric.criteria.map((c) => [c.name, c.weight ?? 1.0]));
+    let weighted_sum = 0;
+    let total_weight = 0;
+    for (const s of scores) {
+        const w = lookup.get(s.name) ?? 1.0;
+        weighted_sum += s.score * w;
+        total_weight += w;
+    }
+    return total_weight > 0 ? weighted_sum / total_weight : 0;
+}
+/** Helper: classify a score into a verdict against the rubric's threshold. */
+export function classify(score, rubric) {
+    const threshold = rubric.satisfied_at ?? 0.85;
+    return score >= threshold ? 'satisfied' : 'needs_revision';
+}
+/** A trivial evaluator useful for tests — accepts the work product as
+ *  satisfying if it matches a predicate. Production code uses an LLM
+ *  evaluator running in a separate context. */
+export function predicateEvaluator(predicate, rubric, feedback_when_failing) {
+    return {
+        async evaluate(work) {
+            const passing = predicate(work);
+            const per_criterion = rubric.criteria.map((c) => ({
+                name: c.name,
+                score: passing ? 1.0 : 0.0,
+                feedback: passing ? 'meets criterion' : feedback_when_failing,
+            }));
+            const overall_score = weightedAverage(per_criterion, rubric);
+            return {
+                per_criterion,
+                overall_score,
+                verdict: passing ? 'satisfied' : 'needs_revision',
+                revision_guidance: passing ? undefined : feedback_when_failing,
+            };
+        },
+    };
+}
+//# sourceMappingURL=outcomes.js.map

dist/outcomes.js.map

+{"version":3,"file":"outcomes.js","sourceRoot":"","sources":["../src/outcomes.ts"],"names":[],"mappings":"AAgGA;;;;;;;;;;;;;;GAcG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,MAAc,EACd,OAAmE,EACnE,SAA0B,EAC1B,OAA8B;IAE9B,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,CAAC,CAAC,CAAI,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAA;IACpE,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAA;IAC3D,MAAM,OAAO,GAA0B,EAAE,CAAA;IACzC,IAAI,QAAQ,GAAkB,IAAI,CAAA;IAClC,IAAI,QAAQ,GAAa,IAAI,CAAA;IAC7B,IAAI,QAAQ,GAA4B,IAAI,CAAA;IAE5C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9B,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;YAC5B,OAAO;gBACL,IAAI,EAAE,aAAa;gBACnB,eAAe,EAAE,CAAC,GAAG,CAAC;gBACtB,UAAU,EAAE,QAAQ;gBACpB,gBAAgB,EAAE,QAAQ;gBAC1B,OAAO;aACR,CAAA;QACH,CAAC;QAED,IAAI,IAAO,CAAA;QACX,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,OAAO,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAA;QACnC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO;gBACL,IAAI,EAAE,QAAQ;gBACd,eAAe,EAAE,CAAC,GAAG,CAAC;gBACtB,UAAU,EAAE,QAAQ;gBACpB,gBAAgB,EAAE,QAAQ;gBAC1B,OAAO;aACR,CAAA;QACH,CAAC;QAED,MAAM,UAAU,GAAG,SAAS,CAAC,IAAI,CAAC,CAAA;QAClC,IAAI,UAA4B,CAAA;QAChC,IAAI,CAAC;YACH,UAAU,GAAG,MAAM,SAAS,CAAC,QAAQ,CAAC,UAAU,EAAE,MAAM,EAAE,CAAC,CAAC,CAAA;QAC9D,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO;gBACL,IAAI,EAAE,QAAQ;gBACd,eAAe,EAAE,CAAC;gBAClB,UAAU,EAAE,IAAI;gBAChB,gBAAgB,EAAE,QAAQ;gBAC1B,OAAO;aACR,CAAA;QACH,CAAC;QAED,MAAM,IAAI,GAAwB;YAChC,SAAS,EAAE,CAAC;YACZ,IAAI;YACJ,eAAe,EAAE,UAAU;YAC3B,UAAU;YACV,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SAC7B,CAAA;QACD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QAClB,QAAQ,GAAG,IAAI,CAAA;QACf,QAAQ,GAAG,UAAU,CAAA;QAErB,IAAI,UAAU,CAAC,OAAO,KAAK,WAAW,EAAE,CAAC;YACvC,OAAO;gBACL,IAAI,EAAE,WAAW;gBACjB,eAAe,EAAE,CAAC;gBAClB,UAAU,EAAE,IAAI;gBAChB,gBAAgB,EAAE,UAAU;gBAC5B,OAAO;aACR,CAAA;QACH,CAAC;QACD,IAAI,UAAU,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;YACpC,OAAO;gBACL,IAAI,EAAE,QAAQ;gBACd,eAAe,EAAE,CAAC;gBAClB,UAAU,EAAE,IAAI;gBAChB,gBAAgB,EAAE,UAAU;gBAC5B,OAAO;aACR,CAAA;QACH,CAAC;QAED,QAAQ;YACN,UAAU,CAAC,iBAAiB,IAAI,kBAAkB,CAAC,UAAU,CAAC,aAAa,CAAC,CAAA;IAChF,CAAC;IAED,OAAO;QACL,IAAI,EAAE,wBAAwB;QAC9B,eAAe,EAAE,GAAG;QACpB,UAAU,EAAE,QAAQ;QACpB,gBAAgB,EAAE,QAAQ;QAC1B,OAAO;KACR,CAAA;AACH,CAAC;AAED;2CAC2C;AAC3C,SAAS,kBAAkB,CAAC,MAAqC;IAC/D,MAAM,OAAO,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;IACzE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAA;IACnC,OAAO,OAAO;SACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,UAAU,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,QAAQ,EAAE,CAAC;SACnE,IAAI,CAAC,IAAI,CAAC,CAAA;AACf,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,eAAe,CAC7B,MAAqC,EACrC,MAAc;IAEd,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAA;IACjC,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,CAAA;IAC7E,IAAI,YAAY,GAAG,CAAC,CAAA;IACpB,IAAI,YAAY,GAAG,CAAC,CAAA;IACpB,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACvB,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,GAAG,CAAA;QACnC,YAAY,IAAI,CAAC,CAAC,KAAK,GAAG,CAAC,CAAA;QAC3B,YAAY,IAAI,CAAC,CAAA;IACnB,CAAC;IACD,OAAO,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC,CAAA;AAC3D,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,QAAQ,CAAC,KAAa,EAAE,MAAc;IACpD,MAAM,SAAS,GAAG,MAAM,CAAC,YAAY,IAAI,IAAI,CAAA;IAC7C,OAAO,KAAK,IAAI,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,gBAAgB,CAAA;AAC5D,CAAC;AAED;;+CAE+C;AAC/C,MAAM,UAAU,kBAAkB,CAChC,SAAoC,EACpC,MAAc,EACd,qBAA6B;IAE7B,OAAO;QACL,KAAK,CAAC,QAAQ,CAAC,IAAY;YACzB,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,CAAA;YAC/B,MAAM,aAAa,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAChD,IAAI,EAAE,CAAC,CAAC,IAAI;gBACZ,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;gBAC1B,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,qBAAqB;aAC9D,CAAC,CAAC,CAAA;YACH,MAAM,aAAa,GAAG,eAAe,CAAC,aAAa,EAAE,MAAM,CAAC,CAAA;YAC5D,OAAO;gBACL,aAAa;gBACb,aAAa;gBACb,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,gBAAgB;gBACjD,iBAAiB,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,qBAAqB;aAC3C,CAAA;QACvB,CAAC;KACF,CAAA;AACH,CAAC"}

dist/vault.d.ts

+import type { ACap, ACapSubject, OSResult } from './types.js';
+/**
+ * Vault — server-side credential injection outside the agent sandbox.
+ *
+ * The pattern adopted from Claude Managed Agents' vault design: credentials
+ * never enter the agent's context. The agent references a credential by
+ * name; the OS resolves and injects it at the chexec boundary; the secret
+ * material is added to the outbound tool request server-side. A
+ * prompt-injection attack that compromises the agent's reasoning cannot
+ * read secrets that never crossed into its context.
+ *
+ * agent-os generalizes CMA's pattern in two ways:
+ *  - bindings are to any ACapSubject, not just MCP server URLs
+ *  - the resolver is pluggable so secrets can live in any backing store
+ *    (file, env, AWS Secrets Manager, Vault, 1Password, on-device keychain)
+ *
+ * v0.1 ships with an in-memory resolver. Production deployments register
+ * a custom resolver via `Vault.withResolver()`.
+ */
+/** A reference name an agent uses to ask for a credential without seeing
+ *  its value. e.g. "factset_api", "sec_edgar_user_agent". */
+export type CredentialRef = string & {
+    readonly __cred_ref: unique symbol;
+};
+/** The actual secret material — must never leave the Vault unrewritten. */
+export type CredentialValue = string & {
+    readonly __cred_value: unique symbol;
+};
+export interface VaultEntry {
+    readonly ref: CredentialRef;
+    /** Which capability subject is allowed to invoke this credential.
+     *  The chexec layer cross-checks the caller's acap against this binding. */
+    readonly bound_to: ACapSubject;
+    /** How the secret material is injected into the tool's outbound request.
+     *  Currently three shapes; extensible. */
+    readonly injection: InjectionShape;
+    /** Optional metadata for audit log entries. Never includes the secret. */
+    readonly description?: string;
+}
+export type InjectionShape = {
+    readonly kind: 'header';
+    readonly name: string;
+    readonly prefix?: string;
+} | {
+    readonly kind: 'query_param';
+    readonly name: string;
+} | {
+    readonly kind: 'bearer_token';
+};
+/** A credential resolver fetches the actual secret value at injection
+ *  time. The Vault holds references to resolvers, not secrets themselves —
+ *  so memory snapshots and core dumps don't leak credentials. */
+export type CredentialResolver = (ref: CredentialRef) => Promise<CredentialValue | null>;
+/** Materialized injection — what the chexec layer should add to the
+ *  outbound tool request. The secret is in `header_value` / `param_value`;
+ *  the agent never sees this struct. */
+export type Injection = {
+    readonly kind: 'header';
+    readonly name: string;
+    readonly header_value: string;
+} | {
+    readonly kind: 'query_param';
+    readonly name: string;
+    readonly param_value: string;
+} | {
+    readonly kind: 'bearer_token';
+    readonly header_value: string;
+};
+/** Audit-loggable record of an injection. Crucially: does NOT include
+ *  the secret value. Records the ref name + binding + outcome. */
+export interface InjectionLog {
+    readonly cred_ref: CredentialRef;
+    readonly caller_acap: string;
+    readonly bound_subject: ACapSubject;
+    readonly injection_kind: InjectionShape['kind'];
+    readonly outcome: 'injected' | 'denied' | 'missing';
+    readonly at: string;
+}
+export declare class Vault {
+    private readonly entries;
+    private resolver;
+    private readonly logs;
+    constructor(resolver?: CredentialResolver);
+    /** Register a credential reference. Only metadata; the secret value
+     *  stays with the resolver. */
+    register(entry: VaultEntry): OSResult<true>;
+    /** Replace the resolver (for plugging in AWS Secrets Manager / Vault /
+     *  1Password etc. without changing the agent contract). */
+    withResolver(resolver: CredentialResolver): void;
+    /**
+     * Inject a credential into a tool call. Returns the materialized
+     * injection or an error. Validates:
+     *   1. The credential exists in the vault.
+     *   2. The caller's acap binds to a subject that matches the
+     *      vault entry's `bound_to`.
+     *   3. The resolver returns a value.
+     *
+     * Logs the outcome regardless of success.
+     */
+    inject(ref: CredentialRef, caller_acap: ACap): Promise<OSResult<Injection>>;
+    /** Read-only snapshot of recent injection log entries. Audit substrate
+     *  consumes this; the agent does not. */
+    recentLogs(limit?: number): ReadonlyArray<InjectionLog>;
+    /** Number of registered entries. */
+    size(): number;
+    /** Archive a credential reference. The entry is removed; resolvers
+     *  may continue to hold the underlying secret independently. */
+    archive(ref: CredentialRef): OSResult<true>;
+    private log;
+}
+/** Helper for tests: build an in-memory resolver pre-populated with a map. */
+export declare function inMemoryResolver(entries: Record<string, string>): CredentialResolver;
+//# sourceMappingURL=vault.d.ts.map

dist/vault.d.ts.map

+{"version":3,"file":"vault.d.ts","sourceRoot":"","sources":["../src/vault.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAA;AAG7D;;;;;;;;;;;;;;;;;GAiBG;AAEH;6DAC6D;AAC7D,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG;IAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,MAAM,CAAA;CAAE,CAAA;AAE3E,2EAA2E;AAC3E,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG;IAAE,QAAQ,CAAC,YAAY,EAAE,OAAO,MAAM,CAAA;CAAE,CAAA;AAE/E,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,GAAG,EAAE,aAAa,CAAA;IAC3B;gFAC4E;IAC5E,QAAQ,CAAC,QAAQ,EAAE,WAAW,CAAA;IAC9B;8CAC0C;IAC1C,QAAQ,CAAC,SAAS,EAAE,cAAc,CAAA;IAClC,0EAA0E;IAC1E,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAC9B;AAED,MAAM,MAAM,cAAc,GACtB;IAAE,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,GAC5E;IAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACvD;IAAE,QAAQ,CAAC,IAAI,EAAE,cAAc,CAAA;CAAE,CAAA;AAErC;;iEAEiE;AACjE,MAAM,MAAM,kBAAkB,GAAG,CAAC,GAAG,EAAE,aAAa,KAAK,OAAO,CAAC,eAAe,GAAG,IAAI,CAAC,CAAA;AAExF;;wCAEwC;AACxC,MAAM,MAAM,SAAS,GACjB;IAAE,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GACjF;IAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;IAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;CAAE,GACrF;IAAE,QAAQ,CAAC,IAAI,EAAE,cAAc,CAAC;IAAC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAAA;AAEpE;kEACkE;AAClE,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAA;IAChC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,QAAQ,CAAC,aAAa,EAAE,WAAW,CAAA;IACnC,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC,MAAM,CAAC,CAAA;IAC/C,QAAQ,CAAC,OAAO,EAAE,UAAU,GAAG,QAAQ,GAAG,SAAS,CAAA;IACnD,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAA;CACpB;AAED,qBAAa,KAAK;IAChB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAuC;IAC/D,OAAO,CAAC,QAAQ,CAAoB;IACpC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAqB;gBAE9B,QAAQ,CAAC,EAAE,kBAAkB;IAIzC;mCAC+B;IAC/B,QAAQ,CAAC,KAAK,EAAE,UAAU,GAAG,QAAQ,CAAC,IAAI,CAAC;IAQ3C;+DAC2D;IAC3D,YAAY,CAAC,QAAQ,EAAE,kBAAkB,GAAG,IAAI;IAIhD;;;;;;;;;OASG;IACG,MAAM,CAAC,GAAG,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,GAAG,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;IA2BjF;6CACyC;IACzC,UAAU,CAAC,KAAK,SAAM,GAAG,aAAa,CAAC,YAAY,CAAC;IAIpD,oCAAoC;IACpC,IAAI,IAAI,MAAM;IAId;oEACgE;IAChE,OAAO,CAAC,GAAG,EAAE,aAAa,GAAG,QAAQ,CAAC,IAAI,CAAC;IAQ3C,OAAO,CAAC,GAAG;CAeZ;AA+CD,8EAA8E;AAC9E,wBAAgB,gBAAgB,CAC9B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAC9B,kBAAkB,CAMpB"}

dist/vault.js

+import { err, ok } from './types.js';
+export class Vault {
+    entries = new Map();
+    resolver;
+    logs = [];
+    constructor(resolver) {
+        this.resolver = resolver ?? makeInMemoryResolver();
+    }
+    /** Register a credential reference. Only metadata; the secret value
+     *  stays with the resolver. */
+    register(entry) {
+        if (this.entries.has(entry.ref)) {
+            return err('capability_denied', `vault entry ${entry.ref} already exists`);
+        }
+        this.entries.set(entry.ref, entry);
+        return ok(true);
+    }
+    /** Replace the resolver (for plugging in AWS Secrets Manager / Vault /
+     *  1Password etc. without changing the agent contract). */
+    withResolver(resolver) {
+        this.resolver = resolver;
+    }
+    /**
+     * Inject a credential into a tool call. Returns the materialized
+     * injection or an error. Validates:
+     *   1. The credential exists in the vault.
+     *   2. The caller's acap binds to a subject that matches the
+     *      vault entry's `bound_to`.
+     *   3. The resolver returns a value.
+     *
+     * Logs the outcome regardless of success.
+     */
+    async inject(ref, caller_acap) {
+        const entry = this.entries.get(ref);
+        if (!entry) {
+            this.log(ref, caller_acap, undefined, 'missing');
+            return err('namespace_violation', `vault has no entry for ${ref}`);
+        }
+        if (!subjectsMatch(caller_acap.subject, entry.bound_to)) {
+            this.log(ref, caller_acap, entry.bound_to, 'denied');
+            return err('capability_denied', `acap subject does not match vault entry binding`, {
+                cred_ref: ref,
+                acap_subject: caller_acap.subject,
+                bound_to: entry.bound_to,
+            });
+        }
+        const value = await this.resolver(ref);
+        if (value === null) {
+            this.log(ref, caller_acap, entry.bound_to, 'missing');
+            return err('namespace_violation', `resolver returned null for ${ref}`);
+        }
+        this.log(ref, caller_acap, entry.bound_to, 'injected');
+        return ok(materialize(entry.injection, value));
+    }
+    /** Read-only snapshot of recent injection log entries. Audit substrate
+     *  consumes this; the agent does not. */
+    recentLogs(limit = 100) {
+        return this.logs.slice(-limit);
+    }
+    /** Number of registered entries. */
+    size() {
+        return this.entries.size;
+    }
+    /** Archive a credential reference. The entry is removed; resolvers
+     *  may continue to hold the underlying secret independently. */
+    archive(ref) {
+        if (!this.entries.has(ref)) {
+            return err('namespace_violation', `vault has no entry for ${ref}`);
+        }
+        this.entries.delete(ref);
+        return ok(true);
+    }
+    log(cred_ref, caller, bound_subject, outcome) {
+        this.logs.push({
+            cred_ref,
+            caller_acap: caller.id,
+            bound_subject: bound_subject ?? caller.subject,
+            injection_kind: 'header',
+            outcome,
+            at: new Date().toISOString(),
+        });
+    }
+}
+function materialize(shape, value) {
+    switch (shape.kind) {
+        case 'header':
+            return {
+                kind: 'header',
+                name: shape.name,
+                header_value: shape.prefix !== undefined ? `${shape.prefix}${value}` : value,
+            };
+        case 'query_param':
+            return { kind: 'query_param', name: shape.name, param_value: value };
+        case 'bearer_token':
+            return { kind: 'bearer_token', header_value: `Bearer ${value}` };
+    }
+}
+function subjectsMatch(a, b) {
+    if (a.kind !== b.kind)
+        return false;
+    switch (a.kind) {
+        case 'tool':
+            return a.name === b.name;
+        case 'mcp_server':
+            return a.server === b.server;
+        case 'resource':
+            return a.uri === b.uri;
+        case 'memory_block':
+            return a.block_id === b.block_id;
+        case 'audit_log':
+            return a.namespace === b.namespace;
+        case 'agent_handoff':
+            return a.target_pattern === b.target_pattern;
+    }
+}
+/** Default resolver — in-memory map for tests and local dev. Production
+ *  deployments register a real resolver via `vault.withResolver()`. */
+function makeInMemoryResolver() {
+    const store = new Map();
+    const fn = async (ref) => store.get(ref) ?? null;
+    fn.set = (ref, value) => store.set(ref, value);
+    return fn;
+}
+/** Helper for tests: build an in-memory resolver pre-populated with a map. */
+export function inMemoryResolver(entries) {
+    const store = new Map(Object.entries(entries));
+    return async (ref) => {
+        const v = store.get(ref);
+        return v !== undefined ? v : null;
+    };
+}
+//# sourceMappingURL=vault.js.map

dist/vault.js.map

+{"version":3,"file":"vault.js","sourceRoot":"","sources":["../src/vault.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,GAAG,EAAE,EAAE,EAAE,MAAM,YAAY,CAAA;AAqEpC,MAAM,OAAO,KAAK;IACC,OAAO,GAAG,IAAI,GAAG,EAA6B,CAAA;IACvD,QAAQ,CAAoB;IACnB,IAAI,GAAmB,EAAE,CAAA;IAE1C,YAAY,QAA6B;QACvC,IAAI,CAAC,QAAQ,GAAG,QAAQ,IAAI,oBAAoB,EAAE,CAAA;IACpD,CAAC;IAED;mCAC+B;IAC/B,QAAQ,CAAC,KAAiB;QACxB,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;YAChC,OAAO,GAAG,CAAC,mBAAmB,EAAE,eAAe,KAAK,CAAC,GAAG,iBAAiB,CAAC,CAAA;QAC5E,CAAC;QACD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,KAAK,CAAC,CAAA;QAClC,OAAO,EAAE,CAAC,IAAI,CAAC,CAAA;IACjB,CAAC;IAED;+DAC2D;IAC3D,YAAY,CAAC,QAA4B;QACvC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAA;IAC1B,CAAC;IAED;;;;;;;;;OASG;IACH,KAAK,CAAC,MAAM,CAAC,GAAkB,EAAE,WAAiB;QAChD,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QACnC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,CAAC,CAAA;YAChD,OAAO,GAAG,CAAC,qBAAqB,EAAE,0BAA0B,GAAG,EAAE,CAAC,CAAA;QACpE,CAAC;QACD,IAAI,CAAC,aAAa,CAAC,WAAW,CAAC,OAAO,EAAE,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACxD,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,EAAE,KAAK,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAA;YACpD,OAAO,GAAG,CACR,mBAAmB,EACnB,iDAAiD,EACjD;gBACE,QAAQ,EAAE,GAAG;gBACb,YAAY,EAAE,WAAW,CAAC,OAAO;gBACjC,QAAQ,EAAE,KAAK,CAAC,QAAQ;aACzB,CACF,CAAA;QACH,CAAC;QACD,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAA;QACtC,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;YACnB,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,EAAE,KAAK,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAA;YACrD,OAAO,GAAG,CAAC,qBAAqB,EAAE,8BAA8B,GAAG,EAAE,CAAC,CAAA;QACxE,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,EAAE,KAAK,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAA;QACtD,OAAO,EAAE,CAAC,WAAW,CAAC,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC,CAAA;IAChD,CAAC;IAED;6CACyC;IACzC,UAAU,CAAC,KAAK,GAAG,GAAG;QACpB,OAAO,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAA;IAChC,CAAC;IAED,oCAAoC;IACpC,IAAI;QACF,OAAO,IAAI,CAAC,OAAO,CAAC,IAAI,CAAA;IAC1B,CAAC;IAED;oEACgE;IAChE,OAAO,CAAC,GAAkB;QACxB,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YAC3B,OAAO,GAAG,CAAC,qBAAqB,EAAE,0BAA0B,GAAG,EAAE,CAAC,CAAA;QACpE,CAAC;QACD,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAA;QACxB,OAAO,EAAE,CAAC,IAAI,CAAC,CAAA;IACjB,CAAC;IAEO,GAAG,CACT,QAAuB,EACvB,MAAY,EACZ,aAAsC,EACtC,OAAgC;QAEhC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;YACb,QAAQ;YACR,WAAW,EAAE,MAAM,CAAC,EAAE;YACtB,aAAa,EAAE,aAAa,IAAI,MAAM,CAAC,OAAO;YAC9C,cAAc,EAAE,QAAQ;YACxB,OAAO;YACP,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SAC7B,CAAC,CAAA;IACJ,CAAC;CACF;AAED,SAAS,WAAW,CAAC,KAAqB,EAAE,KAAsB;IAChE,QAAQ,KAAK,CAAC,IAAI,EAAE,CAAC;QACnB,KAAK,QAAQ;YACX,OAAO;gBACL,IAAI,EAAE,QAAQ;gBACd,IAAI,EAAE,KAAK,CAAC,IAAI;gBAChB,YAAY,EAAE,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,EAAE,CAAC,CAAC,CAAE,KAAgB;aACzF,CAAA;QACH,KAAK,aAAa;YAChB,OAAO,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,WAAW,EAAE,KAAe,EAAE,CAAA;QAChF,KAAK,cAAc;YACjB,OAAO,EAAE,IAAI,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,KAAK,EAAE,EAAE,CAAA;IACpE,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,CAAc,EAAE,CAAc;IACnD,IAAI,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,IAAI;QAAE,OAAO,KAAK,CAAA;IACnC,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;QACf,KAAK,MAAM;YACT,OAAO,CAAC,CAAC,IAAI,KAAM,CAAsB,CAAC,IAAI,CAAA;QAChD,KAAK,YAAY;YACf,OAAO,CAAC,CAAC,MAAM,KAAM,CAAwB,CAAC,MAAM,CAAA;QACtD,KAAK,UAAU;YACb,OAAO,CAAC,CAAC,GAAG,KAAM,CAAqB,CAAC,GAAG,CAAA;QAC7C,KAAK,cAAc;YACjB,OAAO,CAAC,CAAC,QAAQ,KAAM,CAA0B,CAAC,QAAQ,CAAA;QAC5D,KAAK,WAAW;YACd,OAAO,CAAC,CAAC,SAAS,KAAM,CAA2B,CAAC,SAAS,CAAA;QAC/D,KAAK,eAAe;YAClB,OAAO,CAAC,CAAC,cAAc,KAAM,CAAgC,CAAC,cAAc,CAAA;IAChF,CAAC;AACH,CAAC;AAED;uEACuE;AACvE,SAAS,oBAAoB;IAC3B,MAAM,KAAK,GAAG,IAAI,GAAG,EAAkC,CAAA;IACvD,MAAM,EAAE,GAAuB,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,CACnE;IAAC,EAAmF,CAAC,GAAG,GAAG,CAC1F,GAAG,EACH,KAAK,EACL,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAA;IAC1B,OAAO,EAAE,CAAA;AACX,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,gBAAgB,CAC9B,OAA+B;IAE/B,MAAM,KAAK,GAAG,IAAI,GAAG,CAAiB,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAA;IAC9D,OAAO,KAAK,EAAE,GAAG,EAAE,EAAE;QACnB,MAAM,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,GAAa,CAAC,CAAA;QAClC,OAAO,CAAC,KAAK,SAAS,CAAC,CAAC,CAAE,CAAqB,CAAC,CAAC,CAAC,IAAI,CAAA;IACxD,CAAC,CAAA;AACH,CAAC"}

dist/index.d.ts

@@ -5,20 +5,21 @@ /**
  * Permissions, namespaces, resource quotas, content-addressed audit,
- * and downscoped handoff. The OS-level substrate that runs above
+ * downscoped handoff, server-side credential injection, and rubric-
+ * graded self-evaluation. The OS-level substrate that runs above
  * Modal-class sandboxes (Modal, Daytona, RunPod, E2B, local Docker)
  * and below MCP/A2A as the wire formats agents speak.
  *
- * Eight primitives (v0.1 ships the first 5):
+ * Ten primitives (v0.2 ships 7 of 10):
  *
  *   1. spawn(manifest)     ✓ src/spawn.ts
  *   2. acap                ✓ src/acap.ts
- *   3. ns                  — namespaces are typed in src/types.ts;
- *                            enforcement lands in v0.2
+ *   3. ns                  — namespaces typed; enforcement v0.3
  *   4. ulimit-tok          ✓ src/budget.ts
  *   5. chexec              ✓ src/taint.ts
- *   6. audit               — uses @kernel.chat/kbot-finance's hash-chained
- *                            log directly; namespace isolation lands in v0.2
- *   7. handoff             ✓ src/acap.ts (downscope() primitive)
- *   8. snapshot            — content-addressed agent state; v0.2
+ *   6. audit               — uses @kernel.chat/kbot-finance log; ns isolation v0.3
+ *   7. handoff             ✓ src/acap.ts (downscope())
+ *   8. snapshot            — content-addressed agent state; v0.3
+ *   9. vault               ✓ src/vault.ts        (NEW v0.2 — from CMA)
+ *  10. outcomes            ✓ src/outcomes.ts     (NEW v0.2 — from CMA)
  *
- * Apache 2.0. Status: v0.1.0-alpha.0. Reference implementation; not
+ * Apache 2.0. Status: v0.2.0-alpha.0. Reference implementation; not
  * yet certified for multi-tenant production.
@@ -31,2 +32,4 @@ */
 export * from './taint.js';
+export * from './vault.js';
+export * from './outcomes.js';
 //# sourceMappingURL=index.d.ts.map

dist/index.d.ts.map

@@ -1,1 +0,1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,cAAc,YAAY,CAAA;AAC1B,cAAc,WAAW,CAAA;AACzB,cAAc,aAAa,CAAA;AAC3B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,cAAc,YAAY,CAAA;AAC1B,cAAc,WAAW,CAAA;AACzB,cAAc,aAAa,CAAA;AAC3B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA;AAC1B,cAAc,eAAe,CAAA"}

dist/index.js

@@ -5,20 +5,21 @@ /**
  * Permissions, namespaces, resource quotas, content-addressed audit,
- * and downscoped handoff. The OS-level substrate that runs above
+ * downscoped handoff, server-side credential injection, and rubric-
+ * graded self-evaluation. The OS-level substrate that runs above
  * Modal-class sandboxes (Modal, Daytona, RunPod, E2B, local Docker)
  * and below MCP/A2A as the wire formats agents speak.
  *
- * Eight primitives (v0.1 ships the first 5):
+ * Ten primitives (v0.2 ships 7 of 10):
  *
  *   1. spawn(manifest)     ✓ src/spawn.ts
  *   2. acap                ✓ src/acap.ts
- *   3. ns                  — namespaces are typed in src/types.ts;
- *                            enforcement lands in v0.2
+ *   3. ns                  — namespaces typed; enforcement v0.3
  *   4. ulimit-tok          ✓ src/budget.ts
  *   5. chexec              ✓ src/taint.ts
- *   6. audit               — uses @kernel.chat/kbot-finance's hash-chained
- *                            log directly; namespace isolation lands in v0.2
- *   7. handoff             ✓ src/acap.ts (downscope() primitive)
- *   8. snapshot            — content-addressed agent state; v0.2
+ *   6. audit               — uses @kernel.chat/kbot-finance log; ns isolation v0.3
+ *   7. handoff             ✓ src/acap.ts (downscope())
+ *   8. snapshot            — content-addressed agent state; v0.3
+ *   9. vault               ✓ src/vault.ts        (NEW v0.2 — from CMA)
+ *  10. outcomes            ✓ src/outcomes.ts     (NEW v0.2 — from CMA)
  *
- * Apache 2.0. Status: v0.1.0-alpha.0. Reference implementation; not
+ * Apache 2.0. Status: v0.2.0-alpha.0. Reference implementation; not
  * yet certified for multi-tenant production.
@@ -31,2 +32,4 @@ */
 export * from './taint.js';
+export * from './vault.js';
+export * from './outcomes.js';
 //# sourceMappingURL=index.js.map

dist/index.js.map

@@ -1,1 +0,1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,cAAc,YAAY,CAAA;AAC1B,cAAc,WAAW,CAAA;AACzB,cAAc,aAAa,CAAA;AAC3B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,cAAc,YAAY,CAAA;AAC1B,cAAc,WAAW,CAAA;AACzB,cAAc,aAAa,CAAA;AAC3B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA;AAC1B,cAAc,YAAY,CAAA;AAC1B,cAAc,eAAe,CAAA"}

package.json

 {
   "name": "@kernel.chat/agent-os",
-  "version": "0.1.0-alpha.0",
-  "description": "POSIX for AI agents — permissions, namespaces, resource quotas, content-addressed audit, and downscoped handoff. The OS-level substrate that runs above Modal-class sandboxes and below MCP/A2A.",
+  "version": "0.2.0-alpha.0",
+  "description": "POSIX for AI agents — permissions, namespaces, resource quotas, content-addressed audit, downscoped handoff, server-side credential vaults, and rubric-graded outcomes. The OS-level substrate that runs above Modal-class sandboxes and below MCP/A2A.",
   "type": "module",
@@ -25,2 +25,10 @@ "license": "Apache-2.0",
       "types": "./dist/audit.d.ts"
+    },
+    "./vault": {
+      "import": "./dist/vault.js",
+      "types": "./dist/vault.d.ts"
+    },
+    "./outcomes": {
+      "import": "./dist/outcomes.js",
+      "types": "./dist/outcomes.d.ts"
     }
@@ -27,0 +35,0 @@ },

No alert changes

Improved metrics

Total package byte prevSize

107349

44.68%

Number of package files

34

30.77%

Lines of code

1409

54.33%

No dependency changes