edgegate-mcp 0.10.0 → 0.14.0

package/dist/tools/llm_compile.d.ts

@@ -0,0 +1,47 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+export declare const llmCompileInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    /**
+     * Pre-uploaded source artifacts (1 per component). Typical Llama-class:
+     * 3 artifacts (prompt + token + kv_cache). For other architectures pass
+     * M artifacts matching the number of components.
+     */
+    source_artifact_ids: z.ZodArray<z.ZodString, "many">;
+    /** AI Hub device id (e.g. "x_elite", "x2_elite", "sa8295p"). */
+    device_id: z.ZodString;
+    /**
+     * Label only — backend always compiles to QNN_DLC under the hood. The
+     * label is used downstream for profile dispatch hints.
+     */
+    target_runtime: z.ZodDefault<z.ZodString>;
+    /** Default [128, 1] — typical prompt/token split for Llama-class. */
+    sequence_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
+    /** Default [4096]. */
+    context_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
+    /**
+     * Per-component role labels (e.g. ["prompt", "token", "kv_cache"]).
+     * Required when source_artifact_ids has >1 entry — the AI Hub SDK
+     * mandates graph_names for multi-component compiles.
+     */
+    roles: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+}, "strip", z.ZodTypeAny, {
+    workspace_id: string;
+    source_artifact_ids: string[];
+    sequence_lengths: number[];
+    context_lengths: number[];
+    target_runtime: string;
+    device_id: string;
+    roles?: string[] | undefined;
+}, {
+    workspace_id: string;
+    source_artifact_ids: string[];
+    device_id: string;
+    sequence_lengths?: number[] | undefined;
+    context_lengths?: number[] | undefined;
+    roles?: string[] | undefined;
+    target_runtime?: string | undefined;
+}>;
+export type LLMCompileInput = z.infer<typeof llmCompileInputSchema>;
+export declare function llmCompileHandler(client: EdgeGateClient, input: LLMCompileInput): Promise<ToolResult>;

package/dist/tools/llm_compile.js

@@ -0,0 +1,111 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+export const llmCompileInputSchema = z.object({
+    workspace_id: z.string().uuid(),
+    /**
+     * Pre-uploaded source artifacts (1 per component). Typical Llama-class:
+     * 3 artifacts (prompt + token + kv_cache). For other architectures pass
+     * M artifacts matching the number of components.
+     */
+    source_artifact_ids: z.array(z.string().uuid()).min(1).max(8),
+    /** AI Hub device id (e.g. "x_elite", "x2_elite", "sa8295p"). */
+    device_id: z.string().min(1),
+    /**
+     * Label only — backend always compiles to QNN_DLC under the hood. The
+     * label is used downstream for profile dispatch hints.
+     */
+    target_runtime: z.string().min(1).default("genie"),
+    /** Default [128, 1] — typical prompt/token split for Llama-class. */
+    sequence_lengths: z.array(z.number().int().positive()).default([128, 1]),
+    /** Default [4096]. */
+    context_lengths: z.array(z.number().int().positive()).default([4096]),
+    /**
+     * Per-component role labels (e.g. ["prompt", "token", "kv_cache"]).
+     * Required when source_artifact_ids has >1 entry — the AI Hub SDK
+     * mandates graph_names for multi-component compiles.
+     */
+    roles: z.array(z.string().min(1)).optional(),
+});
+export async function llmCompileHandler(client, input) {
+    // Client-side guard: roles required when multi-component. Surface the
+    // same constraint the backend enforces so the agent gets a clean error
+    // without a network round-trip.
+    if (input.source_artifact_ids.length > 1 && (!input.roles || input.roles.length === 0)) {
+        return {
+            isError: true,
+            content: [
+                {
+                    type: "text",
+                    text: `'roles' is required when source_artifact_ids has more than 1 entry ` +
+                        `(the AI Hub SDK requires graph_names for multi-component compiles). ` +
+                        `Typical Llama-class: roles=["prompt", "token", "kv_cache"].`,
+                },
+            ],
+        };
+    }
+    if (input.roles && input.roles.length !== input.source_artifact_ids.length) {
+        return {
+            isError: true,
+            content: [
+                {
+                    type: "text",
+                    text: `'roles' length (${input.roles.length}) must match source_artifact_ids ` +
+                        `length (${input.source_artifact_ids.length}).`,
+                },
+            ],
+        };
+    }
+    try {
+        const job = await client.submitLLMCompile(input.workspace_id, {
+            source_artifact_ids: input.source_artifact_ids,
+            device_id: input.device_id,
+            target_runtime: input.target_runtime,
+            sequence_lengths: input.sequence_lengths,
+            context_lengths: input.context_lengths,
+            ...(input.roles !== undefined ? { roles: input.roles } : {}),
+        });
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: [
+                        `LLM compile job submitted.`,
+                        ``,
+                        `- compile_job_id: ${job.compile_job_id}`,
+                        `- status: ${job.status}`,
+                        `- components: ${input.source_artifact_ids.length}`,
+                        `- device: ${input.device_id}`,
+                        `- target_runtime: ${input.target_runtime} (compiled to QNN_DLC under the hood)`,
+                        ``,
+                        `Poll status with:`,
+                        `  edgegate_check_llm_compile_status({ workspace_id: "${input.workspace_id}", compile_job_id: "${job.compile_job_id}" })`,
+                        ``,
+                        `When status=completed, composite_artifact_id is the QNN_DLC linked model ready to use in a pipeline.`,
+                    ].join("\n"),
+                },
+            ],
+        };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError) {
+            return {
+                isError: true,
+                content: [
+                    {
+                        type: "text",
+                        text: err.status === 402
+                            ? `Your workspace has hit its monthly LLM compile cap (default 100/mo on Pro). ` +
+                                `Upgrade at https://edgegate.frozo.ai/pricing or wait until next billing cycle.\n\n` +
+                                `Detail: ${err.detail}`
+                            : err.status === 401
+                                ? "EDGEGATE_API_KEY is missing, expired, or revoked. Generate a fresh key at " +
+                                    "https://edgegate.frozo.ai/workspace/<id>/settings#api-keys and retry."
+                                : `EdgeGate returned ${err.status}: ${err.detail}`,
+                    },
+                ],
+            };
+        }
+        throw err;
+    }
+}
+//# sourceMappingURL=llm_compile.js.map

package/dist/tools/llm_compile.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm_compile.js","sourceRoot":"","sources":["../../src/tools/llm_compile.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAG7D,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B;;;;OAIG;IACH,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7D,gEAAgE;IAChE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5B;;;OAGG;IACH,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;IAClD,qEAAqE;IACrE,gBAAgB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;IACxE,sBAAsB;IACtB,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC;IACrE;;;;OAIG;IACH,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;CAC7C,CAAC,CAAC;AAIH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,MAAsB,EACtB,KAAsB;IAEtB,sEAAsE;IACtE,uEAAuE;IACvE,gCAAgC;IAChC,IAAI,KAAK,CAAC,mBAAmB,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;QACvF,OAAO;YACL,OAAO,EAAE,IAAI;YACb,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,qEAAqE;wBACrE,sEAAsE;wBACtE,6DAA6D;iBAChE;aACF;SACF,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,mBAAmB,CAAC,MAAM,EAAE,CAAC;QAC3E,OAAO;YACL,OAAO,EAAE,IAAI;YACb,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,mBAAmB,KAAK,CAAC,KAAK,CAAC,MAAM,mCAAmC;wBACxE,WAAW,KAAK,CAAC,mBAAmB,CAAC,MAAM,IAAI;iBAClD;aACF;SACF,CAAC;IACJ,CAAC;IAED,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,gBAAgB,CAAC,KAAK,CAAC,YAAY,EAAE;YAC5D,mBAAmB,EAAE,KAAK,CAAC,mBAAmB;YAC9C,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,cAAc,EAAE,KAAK,CAAC,cAAc;YACpC,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;YACxC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,GAAG,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC7D,CAAC,CAAC;QAEH,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE;wBACJ,4BAA4B;wBAC5B,EAAE;wBACF,qBAAqB,GAAG,CAAC,cAAc,EAAE;wBACzC,aAAa,GAAG,CAAC,MAAM,EAAE;wBACzB,iBAAiB,KAAK,CAAC,mBAAmB,CAAC,MAAM,EAAE;wBACnD,aAAa,KAAK,CAAC,SAAS,EAAE;wBAC9B,qBAAqB,KAAK,CAAC,cAAc,uCAAuC;wBAChF,EAAE;wBACF,mBAAmB;wBACnB,wDAAwD,KAAK,CAAC,YAAY,uBAAuB,GAAG,CAAC,cAAc,MAAM;wBACzH,EAAE;wBACF,sGAAsG;qBACvG,CAAC,IAAI,CAAC,IAAI,CAAC;iBACb;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EACF,GAAG,CAAC,MAAM,KAAK,GAAG;4BAChB,CAAC,CAAC,8EAA8E;gCAC9E,oFAAoF;gCACpF,WAAW,GAAG,CAAC,MAAM,EAAE;4BACzB,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;gCAClB,CAAC,CAAC,4EAA4E;oCAC5E,uEAAuE;gCACzE,CAAC,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE;qBACzD;iBACF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}

package/dist/tools/new_eval_set_version.d.ts

@@ -0,0 +1,18 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+export declare const newEvalSetVersionInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    eval_set_id: z.ZodString;
+    from_version: z.ZodString;
+}, "strict", z.ZodTypeAny, {
+    workspace_id: string;
+    eval_set_id: string;
+    from_version: string;
+}, {
+    workspace_id: string;
+    eval_set_id: string;
+    from_version: string;
+}>;
+export type NewEvalSetVersionInput = z.infer<typeof newEvalSetVersionInputSchema>;
+export declare function newEvalSetVersionHandler(client: EdgeGateClient, input: NewEvalSetVersionInput): Promise<ToolResult>;

package/dist/tools/new_eval_set_version.js

@@ -0,0 +1,47 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+import { formatVersion, surfaceEvalError } from "./create_eval_set.js";
+export const newEvalSetVersionInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+    eval_set_id: z.string().uuid(),
+    from_version: z
+        .string()
+        .uuid()
+        .describe("The version_id of a PUBLISHED version to seed the new draft from."),
+})
+    .strict();
+export async function newEvalSetVersionHandler(client, input) {
+    try {
+        const version = await client.newEvalSetVersion(input.workspace_id, input.eval_set_id, input.from_version);
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: formatVersion(version, `Forked a new draft version seeded from the published cases:`) +
+                        `\n\nEdit it with \`edgegate_update_eval_set\`, then \`edgegate_publish_eval_set\` to re-publish. ` +
+                        `The original published version (and any references / runs bound to its sha) is untouched.`,
+                },
+            ],
+        };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError) {
+            if (err.status === 404) {
+                return {
+                    isError: true,
+                    content: [
+                        {
+                            type: "text",
+                            text: `Unknown eval set or from_version="${input.from_version}". ` +
+                                `Check the ids with \`edgegate_list_eval_sets\`.`,
+                        },
+                    ],
+                };
+            }
+            return surfaceEvalError(err);
+        }
+        throw err;
+    }
+}
+//# sourceMappingURL=new_eval_set_version.js.map

package/dist/tools/new_eval_set_version.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"new_eval_set_version.js","sourceRoot":"","sources":["../../src/tools/new_eval_set_version.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7D,OAAO,EAAE,aAAa,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAEvE,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC;KAC1C,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC9B,YAAY,EAAE,CAAC;SACZ,MAAM,EAAE;SACR,IAAI,EAAE;SACN,QAAQ,CAAC,mEAAmE,CAAC;CACjF,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,MAAsB,EACtB,KAA6B;IAE7B,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,iBAAiB,CAC5C,KAAK,CAAC,YAAY,EAClB,KAAK,CAAC,WAAW,EACjB,KAAK,CAAC,YAAY,CACnB,CAAC;QAEF,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,aAAa,CACX,OAAO,EACP,6DAA6D,CAC9D;wBACD,mGAAmG;wBACnG,2FAA2F;iBAC9F;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACvB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EACF,qCAAqC,KAAK,CAAC,YAAY,KAAK;gCAC5D,iDAAiD;yBACpD;qBACF;iBACF,CAAC;YACJ,CAAC;YACD,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC;QAC/B,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}

package/dist/tools/publish_eval_set.d.ts

@@ -0,0 +1,18 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+export declare const publishEvalSetInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    eval_set_id: z.ZodString;
+    version_id: z.ZodString;
+}, "strict", z.ZodTypeAny, {
+    workspace_id: string;
+    eval_set_id: string;
+    version_id: string;
+}, {
+    workspace_id: string;
+    eval_set_id: string;
+    version_id: string;
+}>;
+export type PublishEvalSetInput = z.infer<typeof publishEvalSetInputSchema>;
+export declare function publishEvalSetHandler(client: EdgeGateClient, input: PublishEvalSetInput): Promise<ToolResult>;

package/dist/tools/publish_eval_set.js

@@ -0,0 +1,74 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+import { formatVersion, surfaceEvalError } from "./create_eval_set.js";
+export const publishEvalSetInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+    eval_set_id: z.string().uuid(),
+    version_id: z.string().uuid(),
+})
+    .strict();
+export async function publishEvalSetHandler(client, input) {
+    try {
+        const version = await client.publishEvalSet(input.workspace_id, input.eval_set_id, input.version_id);
+        const next = `\n\nFeed \`artifact_id\` and \`eval_set_sha256\` into a Behavioral-Gate run ` +
+            `(3b create-bg-run) as the eval-set inputs. The runner recomputes the sha after ` +
+            `download and confirms it matches.`;
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: formatVersion(version, `Published eval set version — validated, hashed, signed, and frozen:`) + next,
+                },
+            ],
+        };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError) {
+            if (err.status === 422) {
+                const violations = extractViolations(err.detail);
+                const body = violations.length > 0
+                    ? `Validation failed with ${violations.length} violation(s) — the version stays a draft:\n` +
+                        violations.map((v) => `  - ${v}`).join("\n")
+                    : `Validation failed (422) — the version stays a draft:\n${err.detail}`;
+                return { isError: true, content: [{ type: "text", text: body }] };
+            }
+            if (err.status === 404) {
+                return {
+                    isError: true,
+                    content: [
+                        {
+                            type: "text",
+                            text: "Unknown eval set or version. Check the eval_set_id / version_id " +
+                                "with `edgegate_list_eval_sets`.",
+                        },
+                    ],
+                };
+            }
+            return surfaceEvalError(err);
+        }
+        throw err;
+    }
+}
+/**
+ * Pull the `violations` array out of a 422 `detail`. The backend returns
+ * `422 {detail: {violations: [...]}}`; the client stringifies `detail`, so
+ * `err.detail` is the JSON-stringified `{violations: [...]}` object (or a
+ * plain string fallback).
+ */
+function extractViolations(detail) {
+    try {
+        const parsed = JSON.parse(detail);
+        if (parsed &&
+            typeof parsed === "object" &&
+            "violations" in parsed &&
+            Array.isArray(parsed.violations)) {
+            return parsed.violations.map((v) => String(v));
+        }
+    }
+    catch {
+        // detail was plain text — fall through
+    }
+    return [];
+}
+//# sourceMappingURL=publish_eval_set.js.map

package/dist/tools/publish_eval_set.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"publish_eval_set.js","sourceRoot":"","sources":["../../src/tools/publish_eval_set.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7D,OAAO,EAAE,aAAa,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAEvE,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC;KACvC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC9B,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;CAC9B,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,MAAsB,EACtB,KAA0B;IAE1B,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,cAAc,CACzC,KAAK,CAAC,YAAY,EAClB,KAAK,CAAC,WAAW,EACjB,KAAK,CAAC,UAAU,CACjB,CAAC;QAEF,MAAM,IAAI,GACR,8EAA8E;YAC9E,iFAAiF;YACjF,mCAAmC,CAAC;QAEtC,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,aAAa,CACX,OAAO,EACP,qEAAqE,CACtE,GAAG,IAAI;iBACX;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACvB,MAAM,UAAU,GAAG,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;gBACjD,MAAM,IAAI,GACR,UAAU,CAAC,MAAM,GAAG,CAAC;oBACnB,CAAC,CAAC,0BAA0B,UAAU,CAAC,MAAM,8CAA8C;wBACzF,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;oBAC9C,CAAC,CAAC,yDAAyD,GAAG,CAAC,MAAM,EAAE,CAAC;gBAC5E,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;YACpE,CAAC;YACD,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACvB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EACF,kEAAkE;gCAClE,iCAAiC;yBACpC;qBACF;iBACF,CAAC;YACJ,CAAC;YACD,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC;QAC/B,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,SAAS,iBAAiB,CAAC,MAAc;IACvC,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAY,CAAC;QAC7C,IACE,MAAM;YACN,OAAO,MAAM,KAAK,QAAQ;YAC1B,YAAY,IAAI,MAAM;YACtB,KAAK,CAAC,OAAO,CAAE,MAAkC,CAAC,UAAU,CAAC,EAC7D,CAAC;YACD,OAAQ,MAAoC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAChF,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,uCAAuC;IACzC,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC"}

package/dist/tools/update_eval_set.d.ts

@@ -0,0 +1,56 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+export declare const updateEvalSetInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    eval_set_id: z.ZodString;
+    version_id: z.ZodString;
+    cases: z.ZodArray<z.ZodObject<{
+        case_id: z.ZodString;
+        prompt: z.ZodString;
+        category: z.ZodEnum<["jailbreak", "forbidden_action", "task", "format"]>;
+        forbidden_actions: z.ZodArray<z.ZodString, "many">;
+        must_refuse: z.ZodBoolean;
+        expected_task_answer: z.ZodNullable<z.ZodString>;
+    }, "strict", z.ZodTypeAny, {
+        prompt: string;
+        case_id: string;
+        category: "jailbreak" | "forbidden_action" | "task" | "format";
+        forbidden_actions: string[];
+        must_refuse: boolean;
+        expected_task_answer: string | null;
+    }, {
+        prompt: string;
+        case_id: string;
+        category: "jailbreak" | "forbidden_action" | "task" | "format";
+        forbidden_actions: string[];
+        must_refuse: boolean;
+        expected_task_answer: string | null;
+    }>, "many">;
+}, "strict", z.ZodTypeAny, {
+    workspace_id: string;
+    cases: {
+        prompt: string;
+        case_id: string;
+        category: "jailbreak" | "forbidden_action" | "task" | "format";
+        forbidden_actions: string[];
+        must_refuse: boolean;
+        expected_task_answer: string | null;
+    }[];
+    eval_set_id: string;
+    version_id: string;
+}, {
+    workspace_id: string;
+    cases: {
+        prompt: string;
+        case_id: string;
+        category: "jailbreak" | "forbidden_action" | "task" | "format";
+        forbidden_actions: string[];
+        must_refuse: boolean;
+        expected_task_answer: string | null;
+    }[];
+    eval_set_id: string;
+    version_id: string;
+}>;
+export type UpdateEvalSetInput = z.infer<typeof updateEvalSetInputSchema>;
+export declare function updateEvalSetHandler(client: EdgeGateClient, input: UpdateEvalSetInput): Promise<ToolResult>;

package/dist/tools/update_eval_set.js

@@ -0,0 +1,60 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+import { evalCaseSchema, formatVersion, surfaceEvalError, } from "./create_eval_set.js";
+export const updateEvalSetInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+    eval_set_id: z.string().uuid(),
+    version_id: z.string().uuid(),
+    cases: z
+        .array(evalCaseSchema)
+        .describe("The FULL replacement list of case dicts for this draft version."),
+})
+    .strict();
+export async function updateEvalSetHandler(client, input) {
+    try {
+        const version = await client.updateEvalSet(input.workspace_id, input.eval_set_id, input.version_id, input.cases);
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: formatVersion(version, `Updated draft version (cases fully replaced):`) +
+                        `\n\nStill a draft — call \`edgegate_publish_eval_set\` to validate, hash, sign, and freeze it.`,
+                },
+            ],
+        };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError) {
+            if (err.status === 409) {
+                return {
+                    isError: true,
+                    content: [
+                        {
+                            type: "text",
+                            text: "This version is already published and is immutable. To change a " +
+                                "published set, fork a fresh draft with " +
+                                `\`edgegate_new_eval_set_version({ workspace_id, eval_set_id, from_version: "${input.version_id}" })\`, ` +
+                                "edit that draft, then publish it.",
+                        },
+                    ],
+                };
+            }
+            if (err.status === 404) {
+                return {
+                    isError: true,
+                    content: [
+                        {
+                            type: "text",
+                            text: "Unknown eval set or version. Check the eval_set_id / version_id " +
+                                "with `edgegate_list_eval_sets`.",
+                        },
+                    ],
+                };
+            }
+            return surfaceEvalError(err);
+        }
+        throw err;
+    }
+}
+//# sourceMappingURL=update_eval_set.js.map

package/dist/tools/update_eval_set.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"update_eval_set.js","sourceRoot":"","sources":["../../src/tools/update_eval_set.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7D,OAAO,EACL,cAAc,EACd,aAAa,EACb,gBAAgB,GACjB,MAAM,sBAAsB,CAAC;AAE9B,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC;KACtC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC9B,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC7B,KAAK,EAAE,CAAC;SACL,KAAK,CAAC,cAAc,CAAC;SACrB,QAAQ,CAAC,iEAAiE,CAAC;CAC/E,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACxC,MAAsB,EACtB,KAAyB;IAEzB,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,aAAa,CACxC,KAAK,CAAC,YAAY,EAClB,KAAK,CAAC,WAAW,EACjB,KAAK,CAAC,UAAU,EAChB,KAAK,CAAC,KAAK,CACZ,CAAC;QAEF,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,aAAa,CAAC,OAAO,EAAE,+CAA+C,CAAC;wBACvE,gGAAgG;iBACnG;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACvB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EACF,kEAAkE;gCAClE,yCAAyC;gCACzC,+EAA+E,KAAK,CAAC,UAAU,UAAU;gCACzG,mCAAmC;yBACtC;qBACF;iBACF,CAAC;YACJ,CAAC;YACD,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACvB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EACF,kEAAkE;gCAClE,iCAAiC;yBACpC;qBACF;iBACF,CAAC;YACJ,CAAC;YACD,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC;QAC/B,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}

package/dist/types.d.ts

@@ -187,6 +187,22 @@ export interface AuditReport {
     generated_at?: string;
 }
 export type HFImportStatus = "queued" | "downloading" | "uploading_to_hub" | "done" | "failed";
+/**
+ * One entry in a multi-part HF import. F-T6.2 (commit 3636c20) added
+ * multi-part `.bin` auto-discovery: a Volko76-style repo with 3 parts
+ * (prompt / token / kv_cache) now unpacks into 3 ArtifactRefs, in addition
+ * to the legacy single `artifact_id` field (which still points at part 1
+ * for backward compat with older MCP clients).
+ */
+export interface ArtifactRef {
+    id: UUID;
+    filename: string;
+    size_bytes: number;
+    /** "prompt" | "token" | "kv_cache" for M=3 multi-part imports; "part_N" otherwise. */
+    role: string;
+    /** 1-based. */
+    part_index: number;
+}
 export interface HFImportJob {
     import_job_id: string;
     status: HFImportStatus;
@@ -196,6 +212,36 @@ export interface HFImportJob {
     size_bytes: number | null;
     filename: string | null;
     error_detail: string | null;
+    /**
+     * Plan B / F-T6.2 — when the HF repo contains multiple .bin parts, this
+     * array surfaces every part (size > 1 for multi-part imports). Omitted
+     * or single-element for legacy single-file imports.
+     */
+    artifacts?: ArtifactRef[];
+}
+export type LLMCompileStatus = "queued" | "running" | "completed" | "failed";
+export interface LLMCompileProgress {
+    compile_jobs_done: number;
+    compile_jobs_total: number;
+}
+/**
+ * Returned by POST /v1/workspaces/{ws}/llm-compile and GET
+ * /v1/workspaces/{ws}/llm-compile/{compile_job_id}. POST returns the
+ * initial queued state; GET reflects current state. On success,
+ * `composite_artifact_id` is the QNN_DLC linked composite ready to be
+ * referenced from a pipeline. On failure, `error_detail` is set.
+ */
+export interface LLMCompileJob {
+    compile_job_id: UUID;
+    status: LLMCompileStatus;
+    composite_artifact_id: UUID | null;
+    error_detail: string | null;
+    progress: LLMCompileProgress;
+    /** Echoed for caller convenience. */
+    device_id?: string;
+    target_runtime?: string;
+    created_at?: string;
+    updated_at?: string;
 }
 export interface MetricDelta {
     current: number | null;
@@ -370,6 +416,153 @@ export interface ByoAuditPage {
     entries: ByoAuditEntry[];
     next_cursor: number | null;
 }
+/**
+ * Balance counts returned at publish and in pack summaries. See the eval-set
+ * authoring contract §4 — the floor is ≥5 must_refuse_with_forbidden cases and
+ * ≥1 task case.
+ */
+export interface EvalBalance {
+    must_refuse: number;
+    must_refuse_with_forbidden: number;
+    task: number;
+    benign: number;
+    total: number;
+}
+/**
+ * One behavioral eval case. Byte-faithful to edgegate/bg/eval_set.py — exactly
+ * six fields. See the eval-set authoring contract §3.
+ */
+export interface EvalCase {
+    case_id: string;
+    prompt: string;
+    category: "jailbreak" | "forbidden_action" | "task" | "format";
+    forbidden_actions: string[];
+    must_refuse: boolean;
+    expected_task_answer: string | null;
+}
+/**
+ * A bundled starter pack a customer can clone from. Returned by
+ * GET /v1/eval-packs.
+ */
+export interface EvalPack {
+    id: string;
+    name: string;
+    case_count: number;
+    balance: EvalBalance;
+}
+/**
+ * An eval-set version. Returned by create / update / publish / new_version.
+ * `eval_set_sha256`, `artifact_id`, `balance`, and `published_at` are null
+ * until the version is published. See the contract §8.
+ */
+export interface EvalSetVersion {
+    id: UUID;
+    eval_set_id: UUID;
+    version: number;
+    status: "draft" | "published";
+    eval_set_sha256: string | null;
+    artifact_id: UUID | null;
+    balance: EvalBalance | null;
+    cases: EvalCase[];
+    created_at: string;
+    published_at: string | null;
+}
+/**
+ * Summary row from GET /v1/workspaces/{ws}/eval-sets. See the contract §7.3.
+ */
+export interface EvalSetSummary {
+    id: UUID;
+    name: string;
+    created_at: string;
+    latest_version: number;
+}
+/**
+ * Body for POST /v1/workspaces/{ws}/reference-captures. Mirrors the backend
+ * `ReferenceCaptureRequest` model (edgegate/api/routes/reference_capture.py).
+ * Exactly one flavor selector — `hf_repo` (auto-FP16) XOR
+ * `reference_upload_artifact_id` (golden) — must be set; else 422.
+ */
+export interface ReferenceCaptureBody {
+    eval_set_artifact_id: UUID;
+    system_prompt: string;
+    decode_config?: Record<string, unknown>;
+    hf_repo?: string;
+    reference_upload_artifact_id?: UUID;
+}
+/**
+ * Response (202) from POST /v1/workspaces/{ws}/reference-captures.
+ */
+export interface ReferenceCaptureJob {
+    job_id: UUID;
+    flavor: string;
+    status: string;
+}
+/**
+ * Response from GET /v1/workspaces/{ws}/reference-captures/{job_id}. The
+ * `reference_artifact_id` (an `ArtifactKind.REFERENCE` artifact) is populated
+ * only when `status` is "done"; feed it into edgegate_create_bg_run.
+ */
+export interface ReferenceCaptureStatus {
+    job_id: UUID;
+    status: string;
+    reference_artifact_id: UUID | null;
+}
+/**
+ * Body for POST /v1/workspaces/{ws}/genie-compile. Mirrors the backend
+ * `GenieCompileRequest` (edgegate/api/routes/compile_genie.py). Exactly one
+ * lane selector — `hf_repo` (Lane A) XOR `onnx_artifact_ids` (Lane B,
+ * genie-ready multi-part) XOR `bundle_artifact_id` (Lane C, precompiled) —
+ * decides the lane; empty/multi-selector → 422.
+ */
+export interface GenieCompileBody {
+    device_label: string;
+    hf_repo?: string;
+    onnx_artifact_ids?: UUID[];
+    bundle_artifact_id?: UUID;
+}
+/**
+ * Response (202) from POST /v1/workspaces/{ws}/genie-compile. Mirrors the
+ * backend `GenieCompileAccepted` schema.
+ */
+export interface GenieCompileJob {
+    job_id: UUID;
+    lane: string;
+    status: string;
+}
+/**
+ * Response from GET /v1/workspaces/{ws}/genie-compile/{job_id}. Mirrors the
+ * backend `GenieCompileStatus` schema. The `bundle_artifact_id` is populated
+ * only when `status` is "done"; feed it into edgegate_create_bg_run.
+ */
+export interface GenieCompileStatus {
+    job_id: UUID;
+    lane: string;
+    status: string;
+    bundle_artifact_id: UUID | null;
+}
+/**
+ * Body for POST /v1/workspaces/{ws}/bg-runs. Mirrors the backend
+ * `BgRunCreateRequest` (edgegate/api/routes/compile_genie.py). Wires a
+ * compiled bundle + eval set + reference into a Run (populates
+ * `Run.runner_config_json`).
+ */
+export interface BgRunCreateBody {
+    bundle_artifact_id: UUID;
+    eval_set_artifact_id: UUID;
+    reference_artifact_id: UUID;
+    vendor?: string;
+    system_prompt?: string;
+    decode_config?: Record<string, unknown>;
+    device_label?: string;
+}
+/**
+ * Response (201) from POST /v1/workspaces/{ws}/bg-runs. Mirrors the backend
+ * `BgRunResponse` schema.
+ */
+export interface BgRunResponse {
+    run_id: UUID;
+    status: string;
+}
 /**
  * Returned by POST /artifacts and POST /artifacts/byo. Mirrors the
  * backend `ArtifactResponse` schema. `storage_url` for BYO artifacts is