@fusionkit/protocol 0.1.0

package/dist/keys.js

@@ -0,0 +1,36 @@
+import { createPrivateKey, createPublicKey, generateKeyPairSync, sign as cryptoSign, verify as cryptoVerify } from "node:crypto";
+import { KEY_ID_HEX_LENGTH } from "./constants.js";
+import { sha256Hex } from "./hash.js";
+export function generateEd25519KeyPair() {
+    const { publicKey, privateKey } = generateKeyPairSync("ed25519");
+    return {
+        publicKeyPem: publicKey.export({ type: "spki", format: "pem" }).toString(),
+        privateKeyPem: privateKey
+            .export({ type: "pkcs8", format: "pem" })
+            .toString()
+    };
+}
+/**
+ * Stable identifier for a public key: the algorithm tag plus a truncated
+ * SHA-256 fingerprint of its PEM. The fingerprint length is a deliberate,
+ * shared constant (KEY_ID_HEX_LENGTH) — 64 bits is ample to identify an
+ * enrolled key while keeping ids short.
+ */
+export function keyIdFromPublicPem(publicKeyPem) {
+    return `ed25519:${sha256Hex(publicKeyPem.trim()).slice(0, KEY_ID_HEX_LENGTH)}`;
+}
+export function signData(privateKeyPem, data) {
+    const key = createPrivateKey(privateKeyPem);
+    const payload = typeof data === "string" ? Buffer.from(data, "utf8") : data;
+    return cryptoSign(null, payload, key).toString("base64");
+}
+export function verifyData(publicKeyPem, data, signatureB64) {
+    const key = createPublicKey(publicKeyPem);
+    const payload = typeof data === "string" ? Buffer.from(data, "utf8") : data;
+    try {
+        return cryptoVerify(null, payload, key, Buffer.from(signatureB64, "base64"));
+    }
+    catch {
+        return false;
+    }
+}

package/dist/model-fusion.d.ts

@@ -0,0 +1,167 @@
+import type { JsonValue } from "./jcs.js";
+export declare const MODEL_FUSION_SCHEMA_NAMES: readonly ["model-call-record.v1", "harness-run-request.v1", "harness-run-result.v1", "harness-candidate-record.v1", "judge-synthesis-record.v1", "benchmark-task-record.v1", "artifact-ref.v1", "tool-call-plan.v1", "tool-execution-record.v1", "ensemble-receipt.v1"];
+export declare const MODEL_FUSION_SCHEMA_BUNDLE_HASH = "sha256:75792f89c091b6ab4fd317a15fb03fd73438563dceff5ccf9f5d7c752dbf35f3";
+export type ModelFusionSchemaName = (typeof MODEL_FUSION_SCHEMA_NAMES)[number];
+export type ModelFusionStatus = "pending" | "running" | "succeeded" | "failed" | "canceled" | "requires_action" | "skipped" | "unsupported";
+export type ModelFusionSideEffects = "none" | "read_only" | "writes_workspace" | "network" | "tool_execution" | "unknown";
+export type ModelFusionHarnessKind = "generic" | "cursor" | "claude_code" | "codex" | "openai_responses";
+export type ModelFusionCapabilityStatus = "supported" | "unsupported" | "degraded" | "unknown";
+export type ModelFusionArtifactKind = "patch" | "log" | "transcript" | "metrics" | "benchmark_task" | "worktree" | "receipt" | "other";
+export type ModelFusionRedactionStatus = "synthetic" | "redacted" | "raw";
+export type ModelFusionErrorKind = "none" | "provider_error" | "validation_error" | "timeout" | "rate_limited" | "tool_denied" | "secret_denied" | "capability_missing" | "internal_error";
+export type ModelFusionChatRole = "system" | "user" | "assistant" | "tool";
+export type BenchmarkTaskKind = "model_fusion" | "harness_coding";
+export type BenchmarkSourceRepo = "fusionkit" | "handoffkit" | "cursorkit" | "mlx-lm";
+export type BenchmarkScorerKind = "exact" | "contains" | "record_join" | "custom";
+export type JudgeSynthesisDecision = "synthesize" | "select_candidate" | "repair_required" | "failed";
+export type ContractMetadataV1<S extends ModelFusionSchemaName> = {
+    schema: S;
+    schema_version: "v1";
+    schema_bundle_hash: string;
+    producer: string;
+    producer_version: string;
+    producer_git_sha: string;
+    created_at: string;
+};
+export type ModelFusionChatMessage = {
+    role: ModelFusionChatRole;
+    content: string;
+};
+export type ModelFusionUsage = {
+    prompt_tokens?: number;
+    completion_tokens?: number;
+    total_tokens?: number;
+};
+export type ModelFusionError = {
+    kind: ModelFusionErrorKind;
+    message?: string;
+    retryable?: boolean;
+};
+export type ArtifactRef = {
+    artifact_id: string;
+    kind: ModelFusionArtifactKind;
+    uri?: string;
+    hash: string;
+    redaction_status?: ModelFusionRedactionStatus;
+};
+export type ArtifactRefV1 = ContractMetadataV1<"artifact-ref.v1"> & ArtifactRef;
+export type ModelCallRecordV1 = ContractMetadataV1<"model-call-record.v1"> & {
+    call_id: string;
+    endpoint_id: string;
+    provider_request_id?: string;
+    model: string;
+    request_hash: string;
+    response_hash?: string;
+    messages: ModelFusionChatMessage[];
+    status: ModelFusionStatus;
+    side_effects: ModelFusionSideEffects;
+    started_at: string;
+    finished_at?: string;
+    latency_ms?: number;
+    usage?: ModelFusionUsage;
+    output_text?: string;
+    error?: ModelFusionError;
+    metadata?: Record<string, JsonValue>;
+};
+export type HarnessRunRequestV1 = ContractMetadataV1<"harness-run-request.v1"> & {
+    request_id: string;
+    harness_kind: ModelFusionHarnessKind;
+    source_repo: string;
+    base_git_sha: string;
+    prompt: string;
+    prompt_hash: string;
+    allowed_tools?: string[];
+    side_effects: ModelFusionSideEffects;
+    requested_capabilities: Record<string, ModelFusionCapabilityStatus>;
+    metadata?: Record<string, JsonValue>;
+};
+export type HarnessRunResultV1 = ContractMetadataV1<"harness-run-result.v1"> & {
+    result_id: string;
+    request_id: string;
+    harness_kind: ModelFusionHarnessKind;
+    status: ModelFusionStatus;
+    candidate_ids: string[];
+    output_summary?: string;
+    artifacts?: ArtifactRef[];
+    capabilities: Record<string, ModelFusionCapabilityStatus>;
+    started_at: string;
+    finished_at?: string;
+    errors?: ModelFusionError[];
+    metadata?: Record<string, JsonValue>;
+};
+export type HarnessCandidateRecordV1 = ContractMetadataV1<"harness-candidate-record.v1"> & {
+    candidate_id: string;
+    request_id: string;
+    harness_kind: ModelFusionHarnessKind;
+    model_call_id?: string;
+    status: ModelFusionStatus;
+    side_effects: ModelFusionSideEffects;
+    branch_name?: string;
+    worktree_path?: string;
+    artifacts?: ArtifactRef[];
+    score?: number;
+    error?: ModelFusionError;
+    metadata?: Record<string, JsonValue>;
+};
+export type JudgeSynthesisRecordV1 = ContractMetadataV1<"judge-synthesis-record.v1"> & {
+    synthesis_id: string;
+    judge_model_call_id?: string;
+    input_candidate_ids: string[];
+    status: ModelFusionStatus;
+    decision: JudgeSynthesisDecision;
+    selected_candidate_id?: string;
+    rationale?: string;
+    final_output: string;
+    score?: number;
+    metrics?: Record<string, JsonValue>;
+};
+export type BenchmarkScorer = {
+    kind: BenchmarkScorerKind;
+    params?: Record<string, JsonValue>;
+};
+export type BenchmarkTaskRecordV1 = ContractMetadataV1<"benchmark-task-record.v1"> & {
+    task_id: string;
+    task_kind: BenchmarkTaskKind;
+    source_repo: BenchmarkSourceRepo;
+    source_sha: string;
+    prompt?: string;
+    prompt_hash: string;
+    setup_hash: string;
+    expected_evidence: string[];
+    scorer: BenchmarkScorer;
+    holdout: boolean;
+    contamination_notes: string;
+    allowed_tools: string[];
+};
+export type ToolCallPlanV1 = ContractMetadataV1<"tool-call-plan.v1"> & {
+    plan_id: string;
+    tool_name: string;
+    arguments_hash: string;
+    side_effects: ModelFusionSideEffects;
+    status: ModelFusionStatus;
+};
+export type ToolExecutionRecordV1 = ContractMetadataV1<"tool-execution-record.v1"> & {
+    execution_id: string;
+    plan_id: string;
+    status: ModelFusionStatus;
+    output_hash?: string;
+    error?: ModelFusionError;
+};
+export type EnsembleReceiptV1 = ContractMetadataV1<"ensemble-receipt.v1"> & {
+    receipt_id: string;
+    run_id: string;
+    status: ModelFusionStatus;
+    artifact_hashes: string[];
+};
+export type ModelFusionRecordV1 = ModelCallRecordV1 | HarnessRunRequestV1 | HarnessRunResultV1 | HarnessCandidateRecordV1 | JudgeSynthesisRecordV1 | BenchmarkTaskRecordV1 | ArtifactRefV1 | ToolCallPlanV1 | ToolExecutionRecordV1 | EnsembleReceiptV1;
+export declare function assertArtifactRefV1(value: unknown): asserts value is ArtifactRefV1;
+export declare function assertModelCallRecordV1(value: unknown): asserts value is ModelCallRecordV1;
+export declare function assertHarnessRunRequestV1(value: unknown): asserts value is HarnessRunRequestV1;
+export declare function assertHarnessRunResultV1(value: unknown): asserts value is HarnessRunResultV1;
+export declare function assertHarnessCandidateRecordV1(value: unknown): asserts value is HarnessCandidateRecordV1;
+export declare function assertJudgeSynthesisRecordV1(value: unknown): asserts value is JudgeSynthesisRecordV1;
+export declare function assertBenchmarkTaskRecordV1(value: unknown): asserts value is BenchmarkTaskRecordV1;
+export declare function assertToolCallPlanV1(value: unknown): asserts value is ToolCallPlanV1;
+export declare function assertToolExecutionRecordV1(value: unknown): asserts value is ToolExecutionRecordV1;
+export declare function assertEnsembleReceiptV1(value: unknown): asserts value is EnsembleReceiptV1;
+export declare function assertModelFusionRecord(value: unknown): asserts value is ModelFusionRecordV1;