edgegate-mcp 0.11.0 → 0.15.0

package/dist/tools/list_eval_sets.js

@@ -0,0 +1,38 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+import { surfaceEvalError } from "./create_eval_set.js";
+export const listEvalSetsInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+})
+    .strict();
+export async function listEvalSetsHandler(client, input) {
+    try {
+        const sets = await client.listEvalSets(input.workspace_id);
+        if (sets.length === 0) {
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: `No eval sets yet in this workspace.\n\n` +
+                            `Create one with \`edgegate_create_eval_set\` (optionally cloning a ` +
+                            `bundled pack from \`edgegate_list_eval_packs\`).`,
+                    },
+                ],
+            };
+        }
+        const header = `Found ${sets.length} eval set(s) in this workspace:\n\n` +
+            `| eval_set_id | name | latest_version | created |\n` +
+            `|---|---|---|---|\n`;
+        const rows = sets
+            .map((s) => `| ${s.id} | ${s.name} | ${s.latest_version} | ${s.created_at.slice(0, 10)} |`)
+            .join("\n");
+        return { content: [{ type: "text", text: header + rows }] };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError)
+            return surfaceEvalError(err);
+        throw err;
+    }
+}
+//# sourceMappingURL=list_eval_sets.js.map

package/dist/tools/list_eval_sets.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"list_eval_sets.js","sourceRoot":"","sources":["../../src/tools/list_eval_sets.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7D,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAExD,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC;KACrC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;CAChC,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,MAAsB,EACtB,KAAwB;IAExB,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,YAAY,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;QAE3D,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtB,OAAO;gBACL,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EACF,yCAAyC;4BACzC,qEAAqE;4BACrE,kDAAkD;qBACrD;iBACF;aACF,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GACV,SAAS,IAAI,CAAC,MAAM,qCAAqC;YACzD,qDAAqD;YACrD,qBAAqB,CAAC;QAExB,MAAM,IAAI,GAAG,IAAI;aACd,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,cAAc,MAAM,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC;aAC1F,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI,EAAE,CAAC,EAAE,CAAC;IAC9D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa;YAAE,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC;QAC/D,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/rerun_bg.d.ts

@@ -0,0 +1,20 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+/**
+ * Re-run an existing behavioral-gate run: clones its already-validated config
+ * (same bundle + eval set + reference + system prompt + device) into a fresh
+ * QUEUED run for the self-hosted runner to pick up. No need to re-supply ids.
+ */
+export declare const rerunBgInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    run_id: z.ZodString;
+}, "strict", z.ZodTypeAny, {
+    run_id: string;
+    workspace_id: string;
+}, {
+    run_id: string;
+    workspace_id: string;
+}>;
+export type RerunBgInput = z.infer<typeof rerunBgInputSchema>;
+export declare function rerunBgHandler(client: EdgeGateClient, input: RerunBgInput): Promise<ToolResult>;

package/dist/tools/rerun_bg.js

@@ -0,0 +1,61 @@
+import { z } from "zod";
+import { EdgeGateError } from "../client.js";
+/**
+ * Re-run an existing behavioral-gate run: clones its already-validated config
+ * (same bundle + eval set + reference + system prompt + device) into a fresh
+ * QUEUED run for the self-hosted runner to pick up. No need to re-supply ids.
+ */
+export const rerunBgInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+    run_id: z
+        .string()
+        .uuid()
+        .describe("An existing behavioral-gate run to re-run (same config, fresh run)."),
+})
+    .strict();
+export async function rerunBgHandler(client, input) {
+    try {
+        const run = await client.rerunBg(input.workspace_id, input.run_id);
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: [
+                        `Re-ran the behavioral-gate run as a new run:`,
+                        ``,
+                        `- run_id: ${run.run_id}`,
+                        `- status: ${run.status}`,
+                        ``,
+                        `Point the self-hosted runner at this new run to execute the gate on-device ` +
+                            `(\`edgegate-runner run --run-id ${run.run_id} --adb-serial <serial>\`).`,
+                    ].join("\n"),
+                },
+            ],
+        };
+    }
+    catch (err) {
+        if (err instanceof EdgeGateError) {
+            return {
+                isError: true,
+                content: [
+                    {
+                        type: "text",
+                        text: err.status === 404
+                            ? "Unknown run or workspace — re-check the ids."
+                            : err.status === 422
+                                ? "That run is not a behavioral-gate run, so it can't be re-run here."
+                                : err.status === 409
+                                    ? "The workspace already has an active run. Wait for it to finish, or cancel " +
+                                        "it with edgegate_cancel_run, then retry."
+                                    : err.status === 403
+                                        ? "You need admin access on this workspace to re-run."
+                                        : `EdgeGate returned ${err.status}: ${err.detail}`,
+                    },
+                ],
+            };
+        }
+        throw err;
+    }
+}
+//# sourceMappingURL=rerun_bg.js.map

package/dist/tools/rerun_bg.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rerun_bg.js","sourceRoot":"","sources":["../../src/tools/rerun_bg.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAG7D;;;;GAIG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC;KAChC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,MAAM,EAAE,CAAC;SACN,MAAM,EAAE;SACR,IAAI,EAAE;SACN,QAAQ,CAAC,qEAAqE,CAAC;CACnF,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,MAAsB,EACtB,KAAmB;IAEnB,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,YAAY,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;QACnE,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE;wBACJ,8CAA8C;wBAC9C,EAAE;wBACF,aAAa,GAAG,CAAC,MAAM,EAAE;wBACzB,aAAa,GAAG,CAAC,MAAM,EAAE;wBACzB,EAAE;wBACF,6EAA6E;4BAC3E,mCAAmC,GAAG,CAAC,MAAM,4BAA4B;qBAC5E,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,8CAA8C;4BAChD,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;gCAClB,CAAC,CAAC,oEAAoE;gCACtE,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;oCAClB,CAAC,CAAC,4EAA4E;wCAC5E,0CAA0C;oCAC5C,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;wCAClB,CAAC,CAAC,oDAAoD;wCACtD,CAAC,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE;qBAC7D;iBACF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}

package/dist/tools/setup_bg_github_action.d.ts

@@ -0,0 +1,44 @@
+import { z } from "zod";
+import { EdgeGateClient } from "../client.js";
+import type { ToolResult } from "./setup_workspace.js";
+/**
+ * Emit everything needed to wire the **Behavioral Gate** into GitHub Actions.
+ * Unlike the standard run gate (server-side, HMAC CI endpoints), BG runs the
+ * quantized model on a REAL device, so it must run on a self-hosted runner with
+ * the device attached, via the `edgegate-bg` composite action. This is a
+ * guidance generator — it composes the workflow + `gh secret` commands + the
+ * self-hosted-runner prerequisites; it makes no API call.
+ */
+export declare const setupBgGithubActionInputSchema: z.ZodObject<{
+    workspace_id: z.ZodString;
+    api_url: z.ZodOptional<z.ZodString>;
+    runner_label: z.ZodOptional<z.ZodString>;
+    bundle_artifact_id: z.ZodOptional<z.ZodString>;
+    eval_set_artifact_id: z.ZodOptional<z.ZodString>;
+    reference_artifact_id: z.ZodOptional<z.ZodString>;
+    adb_serial: z.ZodOptional<z.ZodString>;
+    system_prompt: z.ZodOptional<z.ZodString>;
+    device_label: z.ZodOptional<z.ZodString>;
+}, "strict", z.ZodTypeAny, {
+    workspace_id: string;
+    eval_set_artifact_id?: string | undefined;
+    system_prompt?: string | undefined;
+    bundle_artifact_id?: string | undefined;
+    reference_artifact_id?: string | undefined;
+    device_label?: string | undefined;
+    api_url?: string | undefined;
+    runner_label?: string | undefined;
+    adb_serial?: string | undefined;
+}, {
+    workspace_id: string;
+    eval_set_artifact_id?: string | undefined;
+    system_prompt?: string | undefined;
+    bundle_artifact_id?: string | undefined;
+    reference_artifact_id?: string | undefined;
+    device_label?: string | undefined;
+    api_url?: string | undefined;
+    runner_label?: string | undefined;
+    adb_serial?: string | undefined;
+}>;
+export type SetupBgGithubActionInput = z.infer<typeof setupBgGithubActionInputSchema>;
+export declare function setupBgGithubActionHandler(_client: EdgeGateClient, input: SetupBgGithubActionInput): Promise<ToolResult>;

package/dist/tools/setup_bg_github_action.js

@@ -0,0 +1,109 @@
+import { z } from "zod";
+/**
+ * Emit everything needed to wire the **Behavioral Gate** into GitHub Actions.
+ * Unlike the standard run gate (server-side, HMAC CI endpoints), BG runs the
+ * quantized model on a REAL device, so it must run on a self-hosted runner with
+ * the device attached, via the `edgegate-bg` composite action. This is a
+ * guidance generator — it composes the workflow + `gh secret` commands + the
+ * self-hosted-runner prerequisites; it makes no API call.
+ */
+export const setupBgGithubActionInputSchema = z
+    .object({
+    workspace_id: z.string().uuid(),
+    api_url: z
+        .string()
+        .optional()
+        .describe('EdgeGate API URL. Defaults to "https://edgegateapi.frozo.ai".'),
+    runner_label: z
+        .string()
+        .optional()
+        .describe('Self-hosted runner label. Defaults to "snapdragon".'),
+    bundle_artifact_id: z.string().uuid().optional(),
+    eval_set_artifact_id: z.string().uuid().optional(),
+    reference_artifact_id: z.string().uuid().optional(),
+    adb_serial: z.string().optional().describe("adb serial of the attached device."),
+    system_prompt: z.string().optional(),
+    device_label: z.string().optional(),
+})
+    .strict();
+function shquote(v) {
+    return `'${v.replace(/'/g, "'\\''")}'`;
+}
+export async function setupBgGithubActionHandler(_client, input) {
+    const apiUrl = input.api_url ?? "https://edgegateapi.frozo.ai";
+    const label = input.runner_label ?? "snapdragon";
+    const deviceLabel = input.device_label ?? "Samsung Galaxy S23 Ultra / Snapdragon 8 Gen 2 (SM8550)";
+    const workflow = [
+        `name: Behavioral Gate`,
+        `on:`,
+        `  pull_request:`,
+        `    paths: ['models/**']   # adjust to where your model / quantization lives`,
+        `jobs:`,
+        `  behavioral-gate:`,
+        `    runs-on: [self-hosted, ${label}]`,
+        `    steps:`,
+        `      - uses: actions/checkout@v4`,
+        `      - uses: ./.github/actions/edgegate-bg`,
+        `        with:`,
+        `          api-url: ${apiUrl}`,
+        `          workspace-id: \${{ secrets.EDGEGATE_WORKSPACE_ID }}`,
+        `          token: \${{ secrets.EDGEGATE_TOKEN }}`,
+        `          adb-serial: \${{ secrets.EDGEGATE_ADB_SERIAL }}`,
+        `          bundle-artifact-id: \${{ secrets.EDGEGATE_BUNDLE_ARTIFACT_ID }}`,
+        `          eval-set-artifact-id: \${{ secrets.EDGEGATE_EVAL_SET_ARTIFACT_ID }}`,
+        `          reference-artifact-id: \${{ secrets.EDGEGATE_REFERENCE_ARTIFACT_ID }}`,
+        `          system-prompt: \${{ secrets.EDGEGATE_SYSTEM_PROMPT }}`,
+        `          device-label: ${shquote(deviceLabel)}`,
+    ].join("\n");
+    const secrets = [
+        ["EDGEGATE_WORKSPACE_ID", input.workspace_id],
+        ["EDGEGATE_TOKEN", "<your workspace API key>"],
+        ["EDGEGATE_ADB_SERIAL", input.adb_serial],
+        ["EDGEGATE_BUNDLE_ARTIFACT_ID", input.bundle_artifact_id],
+        ["EDGEGATE_EVAL_SET_ARTIFACT_ID", input.eval_set_artifact_id],
+        ["EDGEGATE_REFERENCE_ARTIFACT_ID", input.reference_artifact_id],
+        ["EDGEGATE_SYSTEM_PROMPT", input.system_prompt],
+    ];
+    const ghCommands = secrets
+        .map(([k, v]) => `gh secret set ${k} -b ${shquote(v ?? `<${k.toLowerCase()}>`)}`)
+        .join("\n");
+    const text = [
+        `## EdgeGate Behavioral Gate — GitHub Actions setup`,
+        ``,
+        `The Behavioral Gate runs the quantized model on a **real device**, so it cannot run on`,
+        `GitHub-hosted runners. It needs a **self-hosted runner with the device attached**, and`,
+        `the runner's exit code is the gate (nonzero on RED → the build fails).`,
+        ``,
+        `**Step 1 — Self-hosted runner.** In your repo: *Settings → Actions → Runners → New`,
+        `self-hosted runner*. Install it on the machine with the device and give it the label`,
+        `\`${label}\`. On that machine install **Python 3.10+** and **Android platform-tools (\`adb\`)**,`,
+        `connect the device over USB, enable USB debugging, and accept the prompt. Verify with`,
+        `\`adb devices\` (must show "device", not "unauthorized").`,
+        ``,
+        `**Step 2 — Copy the action.** Copy the \`.github/actions/edgegate-bg/\` directory`,
+        `(\`action.yml\` + \`create_bg_run.py\`) into your repo — it's self-contained.`,
+        ``,
+        `**Step 3 — Workflow.** Write this to \`.github/workflows/behavioral-gate.yml\`:`,
+        ``,
+        "```yaml",
+        workflow,
+        "```",
+        ``,
+        `**Step 4 — Secrets.** From the repo root:`,
+        ``,
+        "```bash",
+        ghCommands,
+        "```",
+        ``,
+        `Get the artifact ids from \`edgegate_check_genie_compile_status\` (bundle),`,
+        `\`edgegate_publish_eval_set\` (eval-set), and \`edgegate_check_reference_capture_status\``,
+        `(reference). Generate the API key at`,
+        `https://edgegate.frozo.ai/workspace/${input.workspace_id}/settings#api-keys`,
+        ``,
+        `Once pushed, each matching PR runs the gate on the device and fails the build on a`,
+        `behavioral/safety regression. The signed verdict is stored on the run and joins the`,
+        `evidence chain.`,
+    ].join("\n");
+    return { content: [{ type: "text", text }] };
+}
+//# sourceMappingURL=setup_bg_github_action.js.map

package/dist/tools/setup_bg_github_action.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"setup_bg_github_action.js","sourceRoot":"","sources":["../../src/tools/setup_bg_github_action.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,8BAA8B,GAAG,CAAC;KAC5C,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,OAAO,EAAE,CAAC;SACP,MAAM,EAAE;SACR,QAAQ,EAAE;SACV,QAAQ,CAAC,+DAA+D,CAAC;IAC5E,YAAY,EAAE,CAAC;SACZ,MAAM,EAAE;SACR,QAAQ,EAAE;SACV,QAAQ,CAAC,qDAAqD,CAAC;IAClE,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,QAAQ,EAAE;IAChD,oBAAoB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,QAAQ,EAAE;IAClD,qBAAqB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,QAAQ,EAAE;IACnD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,oCAAoC,CAAC;IAChF,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACpC,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CACpC,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,SAAS,OAAO,CAAC,CAAS;IACxB,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC;AACzC,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAC9C,OAAuB,EACvB,KAA+B;IAE/B,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,IAAI,8BAA8B,CAAC;IAC/D,MAAM,KAAK,GAAG,KAAK,CAAC,YAAY,IAAI,YAAY,CAAC;IACjD,MAAM,WAAW,GAAG,KAAK,CAAC,YAAY,IAAI,wDAAwD,CAAC;IAEnG,MAAM,QAAQ,GAAG;QACf,uBAAuB;QACvB,KAAK;QACL,iBAAiB;QACjB,8EAA8E;QAC9E,OAAO;QACP,oBAAoB;QACpB,8BAA8B,KAAK,GAAG;QACtC,YAAY;QACZ,mCAAmC;QACnC,6CAA6C;QAC7C,eAAe;QACf,sBAAsB,MAAM,EAAE;QAC9B,+DAA+D;QAC/D,iDAAiD;QACjD,2DAA2D;QAC3D,2EAA2E;QAC3E,+EAA+E;QAC/E,iFAAiF;QACjF,iEAAiE;QACjE,2BAA2B,OAAO,CAAC,WAAW,CAAC,EAAE;KAClD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,MAAM,OAAO,GAAwC;QACnD,CAAC,uBAAuB,EAAE,KAAK,CAAC,YAAY,CAAC;QAC7C,CAAC,gBAAgB,EAAE,0BAA0B,CAAC;QAC9C,CAAC,qBAAqB,EAAE,KAAK,CAAC,UAAU,CAAC;QACzC,CAAC,6BAA6B,EAAE,KAAK,CAAC,kBAAkB,CAAC;QACzD,CAAC,+BAA+B,EAAE,KAAK,CAAC,oBAAoB,CAAC;QAC7D,CAAC,gCAAgC,EAAE,KAAK,CAAC,qBAAqB,CAAC;QAC/D,CAAC,wBAAwB,EAAE,KAAK,CAAC,aAAa,CAAC;KAChD,CAAC;IACF,MAAM,UAAU,GAAG,OAAO;SACvB,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,OAAO,OAAO,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,WAAW,EAAE,GAAG,CAAC,EAAE,CAAC;SAChF,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,MAAM,IAAI,GAAG;QACX,oDAAoD;QACpD,EAAE;QACF,wFAAwF;QACxF,wFAAwF;QACxF,wEAAwE;QACxE,EAAE;QACF,oFAAoF;QACpF,sFAAsF;QACtF,KAAK,KAAK,wFAAwF;QAClG,uFAAuF;QACvF,2DAA2D;QAC3D,EAAE;QACF,mFAAmF;QACnF,+EAA+E;QAC/E,EAAE;QACF,iFAAiF;QACjF,EAAE;QACF,SAAS;QACT,QAAQ;QACR,KAAK;QACL,EAAE;QACF,2CAA2C;QAC3C,EAAE;QACF,SAAS;QACT,UAAU;QACV,KAAK;QACL,EAAE;QACF,6EAA6E;QAC7E,2FAA2F;QAC3F,sCAAsC;QACtC,uCAAuC,KAAK,CAAC,YAAY,oBAAoB;QAC7E,EAAE;QACF,oFAAoF;QACpF,qFAAqF;QACrF,iBAAiB;KAClB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;AAC/C,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"}