llm-cli-gateway 2.11.0 → 2.12.0

package/dist/upstream-contracts.js

@@ -8,7 +8,7 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "native",
         executable: "vibe-acp",
         entrypointArgs: [],
-        targetVersion: "vibe 2.14.1",
+        targetVersion: "vibe 2.17.1",
         probeArgs: [["--version"], ["--help"]],
         evidence: "Native ACP executable vibe-acp; manual initialize + session/new smoke passed. First runtime pilot.",
         docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.mistral",
@@ -19,9 +19,9 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "native",
         executable: "grok",
         entrypointArgs: ["agent", "stdio"],
-        targetVersion: "grok 0.2.50 (cadf94855)",
+        targetVersion: "grok 0.2.73 (9ff14c43bb)",
         probeArgs: [["agent", "stdio", "--help"]],
-        evidence: "Native ACP via `grok agent stdio`; initialize + session/new smoke passed with isolated leader socket. Second runtime pilot.",
+        evidence: "Native ACP via `grok agent stdio`; initialize + session/new smoke passed with isolated leader socket. Second runtime pilot. Entrypoint re-probed clean at 0.2.73.",
         docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.grok",
     },
     codex: {
@@ -30,10 +30,10 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "adapter_mediated_deferred",
         executable: "codex",
         entrypointArgs: [],
-        targetVersion: "codex-cli 0.139.0",
+        targetVersion: "codex-cli 0.142.4",
         probeArgs: [],
         adapterCandidates: ["zed-industries/codex-acp", "agentclientprotocol/codex-acp"],
-        evidence: "No native ACP entrypoint at codex-cli 0.139.0. Adapter evidence tracked as documentation only; not native gateway ACP support.",
+        evidence: "No native ACP entrypoint at codex-cli 0.142.4. Adapter evidence tracked as documentation only; not native gateway ACP support.",
         docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.codex",
     },
     claude: {
@@ -42,10 +42,10 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "adapter_mediated_deferred",
         executable: "claude",
         entrypointArgs: [],
-        targetVersion: "claude 2.1.175",
+        targetVersion: "claude 2.1.195",
         probeArgs: [],
         adapterCandidates: ["Claude Agent SDK ACP adapter"],
-        evidence: "No native Claude Code CLI ACP entrypoint at claude 2.1.175. Adapter ownership/permission bridging unresolved; deferred.",
+        evidence: "No native Claude Code CLI ACP entrypoint at claude 2.1.195. Adapter ownership/permission bridging unresolved; deferred.",
         docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.claude",
     },
     gemini: {
@@ -54,9 +54,9 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "absent_watchlist",
         executable: "agy",
         entrypointArgs: [],
-        targetVersion: "agy 1.0.7",
+        targetVersion: "agy 1.0.13",
         probeArgs: [],
-        evidence: "agy 1.0.7 has no ACP flag or subcommand. Legacy Gemini CLI ACP evidence does not transfer. Watchlist item.",
+        evidence: "agy 1.0.13 has no ACP flag or subcommand. Legacy Gemini CLI ACP evidence does not transfer. Watchlist item.",
         docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.gemini",
     },
     devin: {
@@ -65,10 +65,10 @@ export const ACP_ENTRYPOINT_CONTRACTS = {
         status: "native",
         executable: "devin",
         entrypointArgs: ["acp"],
-        targetVersion: "devin 2026.5.26-8 (1a388fa9)",
+        targetVersion: "devin 2026.8.18 (16737566)",
         probeArgs: [["--version"]],
         evidence: 'Native ACP entrypoint `devin acp` (stdio JSON-RPC). Slice D1 manual initialize + session/new smoke passed (protocolVersion 1, agent "Affogato", session created). Third native runtime pilot; routing stays config-gated.',
-        docsRef: "docs/plans/devin-integration-scoping.md",
+        docsRef: "docs/plans/first-class-acp-gateway-extension.dag.toml#provider_matrix.devin",
     },
 };
 const PERMISSION_MODES = [
@@ -107,6 +107,7 @@ function subcommand(commandPath, summary, risk, flags = [], options = {}) {
         summary,
         conformanceFixtures: options.fixtures ?? [],
         acknowledgedUpstreamFlags: options.acknowledgedUpstreamFlags ?? [],
+        ...(options.helpProbeExitTolerant ? { helpProbeExitTolerant: true } : {}),
     };
 }
 function acknowledgeSubcommandFlags(subcommands, flags) {
@@ -147,6 +148,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
             agents: subcommand(["agents"], "Inspect and manage Claude agent definitions.", "writes_local_config", [
                 "--add-dir",
                 "--agent",
+                "--all",
                 "--allow-dangerously-skip-permissions",
                 "--cwd",
                 "--dangerously-skip-permissions",
@@ -305,8 +307,11 @@ export const UPSTREAM_CLI_CONTRACTS = {
         acknowledgedUpstreamFlags: [
             "--allow-dangerously-skip-permissions",
             "--allowed",
+            "--ax-screen-reader",
+            "--background",
             "--bare",
             "--betas",
+            "--bg",
             "--brief",
             "--chrome",
             "--dangerously-skip-permissions",
@@ -318,7 +323,6 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "--from-pr",
             "--ide",
             "--include-hook-events",
-            "--mcp-debug",
             "--name",
             "--no-chrome",
             "--plugin-dir",
@@ -329,6 +333,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "--remote-control-session-name-prefix",
             "--replay-user-messages",
             "--resume",
+            "--safe-mode",
             "--tmux",
             "--version",
             "--worktree",
@@ -416,6 +421,12 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 ],
                 expect: "pass",
             },
+            {
+                id: "claude-background-acknowledged-not-emitted",
+                description: "Claude 2.1.195 advertises --bg/--background (background agent), but the gateway acknowledges them without emitting; caller argv is rejected",
+                args: ["-p", "hello", "--background"],
+                expect: "fail",
+            },
         ],
     },
     codex: {
@@ -471,33 +482,40 @@ export const UPSTREAM_CLI_CONTRACTS = {
             ], {
                 children: {
                     resume: subcommand(["exec", "resume"], "Resume Codex sessions from the interactive CLI.", "executes_agent", [
-                        "--add-dir",
                         "--all",
-                        "--cd",
                         "--config",
                         "--dangerously-bypass-approvals-and-sandbox",
                         "--dangerously-bypass-hook-trust",
                         "--disable",
                         "--enable",
+                        "--ephemeral",
+                        "--ignore-rules",
+                        "--ignore-user-config",
                         "--image",
-                        "--include-non-interactive",
+                        "--json",
                         "--last",
-                        "--local-provider",
                         "--model",
-                        "--no-alt-screen",
-                        "--oss",
-                        "--profile",
-                        "--remote",
-                        "--remote-auth-token-env",
-                        "--sandbox",
+                        "--output-last-message",
+                        "--output-schema",
+                        "--skip-git-repo-check",
                         "--strict-config",
                     ]),
                     review: subcommand(["exec", "review"], "Run Codex code review workflows.", "executes_agent", [
                         "--base",
                         "--commit",
                         "--config",
+                        "--dangerously-bypass-approvals-and-sandbox",
+                        "--dangerously-bypass-hook-trust",
                         "--disable",
                         "--enable",
+                        "--ephemeral",
+                        "--ignore-rules",
+                        "--ignore-user-config",
+                        "--json",
+                        "--model",
+                        "--output-last-message",
+                        "--output-schema",
+                        "--skip-git-repo-check",
                         "--strict-config",
                         "--title",
                         "--uncommitted",
@@ -889,7 +907,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
             plugins: subcommand(["plugins"], "Alias for Antigravity plugin management.", "writes_local_config", [], {
                 aliases: ["plugin"],
             }),
-            update: subcommand(["update"], "Update Antigravity CLI to the current release.", "writes_local_config", [], { tokenCost: "small" }),
+            update: subcommand(["update"], "Update Antigravity CLI to the current release.", "writes_local_config", [], { tokenCost: "small", helpProbeExitTolerant: true }),
         },
         maxPositionals: 1,
         mcpTools: ["gemini_request", "gemini_request_async"],
@@ -911,6 +929,8 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "attachments",
             "skipTrust",
             "yolo",
+            "project",
+            "newProject",
         ],
         flags: {
             "--print": { arity: "none", description: "Run a single prompt non-interactively" },
@@ -926,14 +946,13 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "--conversation": { arity: "one", description: "Resume a previous conversation by ID" },
             "--continue": { arity: "none", description: "Continue the most recent conversation" },
             "-c": { arity: "none", description: "Short alias for --continue" },
+            "--project": { arity: "one", description: "Antigravity project ID for this session" },
+            "--new-project": {
+                arity: "none",
+                description: "Create a new Antigravity project for this session",
+            },
         },
-        acknowledgedUpstreamFlags: [
-            "--prompt-interactive",
-            "-i",
-            "--print-timeout",
-            "--log-file",
-            "--version",
-        ],
+        acknowledgedUpstreamFlags: ["--log-file", "--print-timeout", "--prompt-interactive"],
         env: {},
         conformanceFixtures: [
             {
@@ -972,6 +991,18 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 args: ["--print", "hello", "-o", "json"],
                 expect: "fail",
             },
+            {
+                id: "gemini-project-wired",
+                description: "Antigravity 1.0.13: --project <ID> is wired",
+                args: ["--print", "hello", "--project", "proj-123"],
+                expect: "pass",
+            },
+            {
+                id: "gemini-new-project-wired",
+                description: "Antigravity 1.0.13: --new-project is wired",
+                args: ["--print", "hello", "--new-project"],
+                expect: "pass",
+            },
         ],
     },
     grok: {
@@ -1022,6 +1053,17 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 },
             }),
             completions: subcommand(["completions"], "Generate Grok shell completions.", "read_only", ["--leader-socket"], { tier: "inspect" }),
+            dashboard: subcommand(["dashboard"], "Open the Agent Dashboard view at startup.", "read_only", ["--leader-socket"], {
+                tier: "inspect",
+                fixtures: [
+                    {
+                        id: "grok-dashboard",
+                        description: "grok dashboard subcommand (leader socket passthrough)",
+                        args: ["--leader-socket", "/tmp/dash.sock"],
+                        expect: "pass",
+                    },
+                ],
+            }),
             export: subcommand(["export"], "Export Grok session data.", "read_only", ["--clipboard", "--leader-socket"], { tier: "inspect" }),
             import: subcommand(["import"], "Import Grok session data.", "writes_local_config", [
                 "--json",
@@ -1052,7 +1094,49 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 tier: "inspect",
             }),
             setup: subcommand(["setup"], "Configure Grok CLI local setup.", "writes_local_config", ["--leader-socket"], { exposure: "not_exposed" }),
-            ssh: subcommand(["ssh"], "Manage Grok SSH integration.", "network", ["--leader-socket"]),
+            ssh: subcommand(["ssh"], "Manage Grok SSH integration.", "network", ["--leader-socket"], {
+                acknowledgedUpstreamFlags: [
+                    "--agent",
+                    "--agents",
+                    "--allow",
+                    "--always-approve",
+                    "--best-of-n",
+                    "--check",
+                    "--continue",
+                    "--cwd",
+                    "--deny",
+                    "--disable-web-search",
+                    "--disallowed-tools",
+                    "--effort",
+                    "--experimental-memory",
+                    "--fork-session",
+                    "--json-schema",
+                    "--max-turns",
+                    "--model",
+                    "--no-alt-screen",
+                    "--no-memory",
+                    "--no-plan",
+                    "--no-subagents",
+                    "--oauth",
+                    "--output-format",
+                    "--permission-mode",
+                    "--prompt-file",
+                    "--prompt-json",
+                    "--reasoning-effort",
+                    "--restore-code",
+                    "--resume",
+                    "--rules",
+                    "--sandbox",
+                    "--session-id",
+                    "--single",
+                    "--system-prompt-override",
+                    "--tools",
+                    "--verbatim",
+                    "--version",
+                    "--worktree",
+                    "--worktree-ref",
+                ],
+            }),
             trace: subcommand(["trace"], "Inspect Grok trace data.", "read_only", ["--json", "--leader-socket", "--local", "--output"], { tier: "diagnostic" }),
             update: subcommand(["update"], "Update the Grok CLI binary.", "updates_binary", [
                 "--alpha",
@@ -1067,7 +1151,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
             worktree: subcommand(["worktree"], "Manage Grok worktree sessions.", "writes_local_config", ["--leader-socket"]),
         }, GROK_DEBUG_HELP_FLAGS),
         maxPositionals: 0,
-        acknowledgedUpstreamFlags: GROK_DEBUG_HELP_FLAGS,
+        acknowledgedUpstreamFlags: [...GROK_DEBUG_HELP_FLAGS, "--session-id"],
         mcpTools: ["grok_request", "grok_request_async"],
         mcpParameters: [
             "prompt",
@@ -1112,6 +1196,9 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "restoreCode",
             "leaderSocket",
             "nativeWorktree",
+            "worktreeRef",
+            "forkSession",
+            "jsonSchema",
         ],
         flags: {
             "-p": { arity: "one", description: "Prompt text" },
@@ -1197,22 +1284,40 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 description: "Custom leader socket path (isolated leader, Grok 0.2.32+)",
             },
             "--single": { arity: "one", description: "Single-turn prompt" },
-            "--todo-gate": { arity: "none", description: "Enable runtime turn-end TodoGate" },
+            "--todo-gate": {
+                arity: "none",
+                description: "Enable runtime turn-end TodoGate (accepted at 0.2.60+ but hidden from --help)",
+                hiddenFromHelp: true,
+            },
             "--verbatim": { arity: "none", description: "Send prompt exactly as given" },
             "--version": { arity: "none", description: "Print version" },
             "--worktree": {
                 arity: "optional",
                 description: "Start the session in a new git worktree, optionally named",
             },
+            "--worktree-ref": {
+                arity: "one",
+                description: "Git ref to base the worktree on (requires --worktree)",
+            },
+            "--fork-session": {
+                arity: "none",
+                description: "Fork the resumed session into a new session ID",
+            },
+            "--json-schema": {
+                arity: "one",
+                description: "JSON Schema literal constraining structured output (implies json output)",
+            },
             "--compaction-mode": {
                 arity: "one",
                 values: ["summary", "transcript", "segments"],
                 description: "Compaction mode (default summary; sets GROK_COMPACTION_MODE). `segments` persists per-segment markdown.",
+                hiddenFromHelp: true,
             },
             "--compaction-detail": {
                 arity: "one",
                 values: ["none", "minimal", "balanced", "verbose"],
                 description: "Segment verbatim detail (default verbose; sets GROK_COMPACTION_DETAIL). Only affects `--compaction-mode segments`.",
+                hiddenFromHelp: true,
             },
         },
         env: {},
@@ -1370,6 +1475,35 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 args: ["-p", "hello", "--agent", "reviewer", "--leader-socket", "/tmp/leader.sock"],
                 expect: "pass",
             },
+            {
+                id: "grok-json-schema-wired",
+                description: "Grok 0.2.73: --json-schema <SCHEMA> is wired (structured output, implies json)",
+                args: [
+                    "-p",
+                    "hello",
+                    "--json-schema",
+                    '{"type":"object","properties":{"name":{"type":"string"}}}',
+                ],
+                expect: "pass",
+            },
+            {
+                id: "grok-fork-session-wired",
+                description: "Grok 0.2.73: --fork-session is wired (fork resumed session into a new ID)",
+                args: ["-p", "hello", "--resume", "sess-1", "--fork-session"],
+                expect: "pass",
+            },
+            {
+                id: "grok-worktree-ref-wired",
+                description: "Grok 0.2.73: --worktree-ref <REF> is wired (with --worktree)",
+                args: ["-p", "hello", "--worktree", "--worktree-ref", "main"],
+                expect: "pass",
+            },
+            {
+                id: "grok-session-id-acknowledged-not-emitted",
+                description: "Grok 0.2.73 advertises --session-id, but the gateway owns session-id lifecycle and does not emit it; caller argv is rejected",
+                args: ["-p", "hello", "--session-id", "11111111-1111-1111-1111-111111111111"],
+                expect: "fail",
+            },
         ],
     },
     mistral: {
@@ -1377,7 +1511,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
         executable: "vibe",
         upstream: "Mistral Vibe CLI",
         upstreamMetadata: {
-            sourceUrls: ["https://github.com/mistralai/mistral-vibe/releases"],
+            sourceUrls: ["https://api.github.com/repos/mistralai/mistral-vibe/releases/latest"],
             packageName: "mistral-vibe",
             repo: "https://github.com/mistralai/mistral-vibe",
             installDocsUrl: "https://github.com/mistralai/mistral-vibe#installation",
@@ -1459,7 +1593,7 @@ export const UPSTREAM_CLI_CONTRACTS = {
                 description: "Additional writable workspace directory (Phase 4 slice ζ; repeat once per directory)",
             },
         },
-        acknowledgedUpstreamFlags: ["--auto-approve"],
+        acknowledgedUpstreamFlags: ["--auto-approve", "--check-upgrade", "--yolo"],
         env: {
             VIBE_ACTIVE_MODEL: {
                 arity: "one",
@@ -1558,11 +1692,25 @@ export const UPSTREAM_CLI_CONTRACTS = {
             },
             {
                 id: "mistral-current-help-surface",
-                description: "Vibe 2.12.x help surface: --prompt, -v, --version, --setup accepted",
+                description: "Vibe 2.17.1 request-time help surface: --prompt, -v, --version, --setup accepted",
                 args: ["--prompt", "hello", "--agent", "auto-approve", "-v", "--version", "--setup"],
                 env: { VIBE_ACTIVE_MODEL: "mistral-medium-3.5" },
                 expect: "pass",
             },
+            {
+                id: "mistral-yolo-shortcut-rejected",
+                description: "Vibe 2.17.1 advertises --yolo as a shortcut, but the gateway keeps using explicit --agent auto-approve",
+                args: ["-p", "hello", "--yolo"],
+                env: { VIBE_ACTIVE_MODEL: "mistral-medium-3.5" },
+                expect: "fail",
+            },
+            {
+                id: "mistral-check-upgrade-rejected",
+                description: "Vibe 2.17.1 advertises --check-upgrade, but gateway request validation rejects update-prompt flags",
+                args: ["--check-upgrade"],
+                env: { VIBE_ACTIVE_MODEL: "mistral-medium-3.5" },
+                expect: "fail",
+            },
             {
                 id: "mistral-resume-bare",
                 description: "Vibe --resume without session ID is accepted (optional arity)",
@@ -1604,13 +1752,22 @@ export const UPSTREAM_CLI_CONTRACTS = {
             "--model": { arity: "one", description: "AI model for this session" },
             "--permission-mode": {
                 arity: "one",
-                values: ["normal", "auto", "dangerous", "yolo", "bypass"],
-                description: "Permission mode (normal/auto = read-only auto-approve; dangerous/yolo/bypass = approve all)",
+                values: ["auto", "smart", "dangerous"],
+                description: "Permission mode (auto = read-only auto-approve; smart = additionally auto-runs safe actions per fast model; dangerous = approve all)",
             },
             "--prompt-file": { arity: "one", description: "Load the initial prompt from a file" },
             "--resume": { arity: "one", description: "Resume a specific session by ID" },
             "--continue": { arity: "none", description: "Resume the most recent session in cwd" },
         },
+        acknowledgedUpstreamFlags: [
+            "--agent-config",
+            "--config",
+            "--export",
+            "--print",
+            "--respect-workspace-trust",
+            "--sandbox",
+            "--version",
+        ],
         env: {},
         conformanceFixtures: [
             {
@@ -1633,20 +1790,20 @@ export const UPSTREAM_CLI_CONTRACTS = {
             },
             {
                 id: "devin-permission-mode",
-                description: "Valid --permission-mode accepted (dangerous alias bypass)",
-                args: ["-p", "hello", "--permission-mode", "bypass"],
+                description: "Valid --permission-mode 'dangerous' accepted",
+                args: ["-p", "hello", "--permission-mode", "dangerous"],
                 expect: "pass",
             },
             {
                 id: "devin-permission-mode-auto",
-                description: "Valid --permission-mode alias 'auto' (= normal) accepted",
+                description: "Valid --permission-mode 'auto' accepted",
                 args: ["-p", "hello", "--permission-mode", "auto"],
                 expect: "pass",
             },
             {
-                id: "devin-permission-mode-yolo",
-                description: "Valid --permission-mode alias 'yolo' (= dangerous) accepted",
-                args: ["-p", "hello", "--permission-mode", "yolo"],
+                id: "devin-permission-mode-smart",
+                description: "Valid --permission-mode 'smart' accepted",
+                args: ["-p", "hello", "--permission-mode", "smart"],
                 expect: "pass",
             },
             {
@@ -2224,7 +2381,7 @@ function probeInstalledCliSubcommands(cli, timeoutMs) {
                 break;
             }
             outputs.push(`${result.stdout ?? ""}\n${result.stderr ?? ""}`);
-            if (result.status !== 0) {
+            if (result.status !== 0 && !sub.helpProbeExitTolerant) {
                 warnings.push(`${contract.executable} ${args.join(" ")} exited with status ${result.status}`);
             }
         }

package/dist/validation-orchestrator.d.ts

@@ -3,11 +3,13 @@ import { type ProviderRuntimeStatus } from "./provider-status.js";
 import type { ApiProviderRuntime } from "./config.js";
 import { type NormalizedValidationResult, type ValidationProvider } from "./validation-normalizer.js";
 import { type ValidationReport } from "./validation-report.js";
+import type { ValidationRunStore } from "./job-store.js";
 import { type ValidationIntent } from "./validation-prompts.js";
 export interface ValidationOrchestratorDeps {
     asyncJobManager: AsyncJobManager;
     getProviderRuntimeStatus?: (provider: ValidationProvider) => ProviderRuntimeStatus;
     apiProviders?: ApiProviderRuntime[];
+    validationRunStore?: ValidationRunStore | null;
 }
 export interface StartValidationInput {
     intent: ValidationIntent;
@@ -21,7 +23,7 @@ export interface StartValidationInput {
 export interface ValidationRunReport {
     success: boolean;
     validationId: string;
-    status: "running" | "partial" | "not_started";
+    status: "running" | "partial" | "not_started" | "completed";
     startedAt: string;
     intent: ValidationIntent;
     originalRequest: {
@@ -32,7 +34,7 @@ export interface ValidationRunReport {
     modelList: ValidationProvider[];
     results: NormalizedValidationResult[];
     synthesis: {
-        status: "not_requested" | "waiting_for_provider_results" | "running" | "skipped";
+        status: "not_requested" | "waiting_for_provider_results" | "running" | "skipped" | "completed";
         judgeModel: ValidationProvider | null;
         rawJobReference: NormalizedValidationResult["rawJobReference"];
         note: string;
@@ -45,5 +47,6 @@ export declare function startJudgeSynthesis(deps: ValidationOrchestratorDeps, in
     question: string;
     providerResults: NormalizedValidationResult[];
     judgeProvider: ValidationProvider;
+    validationId?: string;
 }): ValidationRunReport["synthesis"];
 export declare function collectValidationJobResult(deps: ValidationOrchestratorDeps, provider: ValidationProvider, jobId: string, model: string | null, maxChars?: number): NormalizedValidationResult | null;

package/dist/validation-orchestrator.js

@@ -3,7 +3,8 @@ import { getProviderRuntimeStatus } from "./provider-status.js";
 import { createApiProvider } from "./api-provider.js";
 import { prepareApiRequest } from "./api-request.js";
 import { normalizeJobResult, normalizeSkippedProvider, normalizeStartedJob, } from "./validation-normalizer.js";
-import { buildValidationReport } from "./validation-report.js";
+import { buildValidationReport, deriveValidationRunStatus, } from "./validation-report.js";
+import { getRequestContext, principalCanAccess, resolveOwnerPrincipal } from "./request-context.js";
 import { buildJudgePrompt, buildValidationPrompt, } from "./validation-prompts.js";
 function findApiReviewer(deps, provider) {
     return deps.apiProviders?.find(p => p.name === provider) ?? null;
@@ -21,8 +22,13 @@ function dispatchProviderJob(deps, provider, prompt, correlationId) {
     if (api) {
         const apiProvider = createApiProvider(api.name, api.kind);
         const apiRequest = prepareApiRequest(api, { prompt });
-        return deps.asyncJobManager.startHttpJob({ provider: apiProvider, apiRequest, correlationId })
-            .snapshot;
+        return deps.asyncJobManager.startHttpJob({
+            provider: apiProvider,
+            apiRequest,
+            correlationId,
+            writeFlightStart: true,
+            flightRecorderEntry: { model: apiRequest.model, prompt },
+        }).snapshot;
     }
     return deps.asyncJobManager.startJob(provider, buildProviderArgs(provider, prompt), correlationId);
 }
@@ -39,9 +45,16 @@ export function startValidationRun(deps, input) {
     const providers = uniqueProviders(input.providers);
     const results = providers.map(provider => startProviderJob(deps, provider, prompt, validationId));
     const runningCount = results.filter(result => result.status === "running").length;
-    const skippedCount = results.filter(result => result.status === "skipped").length;
     const synthesis = plannedJudgeSynthesis(input);
-    const status = runningCount === 0 ? "not_started" : skippedCount > 0 ? "partial" : "running";
+    const status = deriveValidationRunStatus(results, synthesis.status);
+    persistValidationRun(deps, {
+        validationId,
+        startedAt,
+        intent: input.intent,
+        input,
+        providers,
+        results,
+    });
     const reportInput = {
         validationId,
         status,
@@ -103,6 +116,7 @@ export function startJudgeSynthesis(deps, input) {
         question: input.question,
         providerResults: completedResults,
     }), `validation-judge-${randomUUID()}-${input.judgeProvider}`);
+    linkJudgeJob(deps, input.validationId, input.judgeProvider, snapshot);
     return {
         status: "running",
         judgeModel: input.judgeProvider,
@@ -150,6 +164,58 @@ function plannedJudgeSynthesis(input) {
         note: "Collect provider results first, then call synthesize_validation with those results.",
     };
 }
+function linkJudgeJob(deps, validationId, provider, snapshot) {
+    const store = deps.validationRunStore;
+    if (!store || !validationId)
+        return;
+    try {
+        const run = store.getValidationRun(validationId);
+        if (!run)
+            return;
+        if (!principalCanAccess(run.ownerPrincipal, resolveOwnerPrincipal(getRequestContext())))
+            return;
+        store.setValidationJudgeLink(validationId, {
+            provider: String(provider),
+            jobId: snapshot.id,
+            correlationId: snapshot.correlationId,
+        });
+    }
+    catch {
+    }
+}
+function persistValidationRun(deps, args) {
+    const store = deps.validationRunStore;
+    if (!store)
+        return;
+    try {
+        const providerLinks = args.results
+            .filter(result => result.rawJobReference !== null)
+            .map(result => ({
+            provider: String(result.provider),
+            jobId: result.rawJobReference.jobId,
+            correlationId: result.rawJobReference.correlationId,
+        }));
+        store.recordValidationRun({
+            validationId: args.validationId,
+            ownerPrincipal: resolveOwnerPrincipal(getRequestContext()),
+            intent: args.intent,
+            createdAt: args.startedAt,
+            requestJson: JSON.stringify({
+                question: args.input.question,
+                content: args.input.content,
+                focus: args.input.focus,
+                riskLevel: args.input.riskLevel,
+                modelList: args.providers,
+                judgeProvider: args.input.judgeProvider ?? null,
+            }),
+            providerLinks,
+            judgeLink: null,
+            status: "running",
+        });
+    }
+    catch {
+    }
+}
 function buildProviderArgs(provider, prompt) {
     if (provider === "claude" || provider === "grok" || provider === "mistral") {
         return ["-p", prompt];