@martinloop/mcp 0.8.4 → 0.8.6

package/dist/arcade/capabilities.d.ts

@@ -54,7 +54,7 @@ export declare function buildArcadeToolDefinitions(capabilities: unknown): reado
             readonly displayOutcome: {
                 readonly anyOf: readonly [{
                     readonly type: "string";
-                    readonly enum: readonly ["VERIFIED", "STOPPED", "NEEDS REVIEW"];
+                    readonly enum: readonly ["VERIFIED", "STOPPED", "NEEDS_REVIEW"];
                 }, {
                     readonly type: "null";
                 }];

package/dist/arcade/capabilities.js

@@ -1,3 +1,4 @@
+import { VERIFIED_HANDOFF_OUTCOMES } from "../vendor/contracts/index.js";
 export const MARTIN_ARCADE_UI_EXTENSION_ID = "io.modelcontextprotocol/ui";
 export const MARTIN_ARCADE_MIME_TYPE = "text/html;profile=mcp-app";
 export const MARTIN_ARCADE_RESOURCE_URI = "ui://martinloop/arcade/index.html";
@@ -51,7 +52,7 @@ export function buildArcadeToolDefinitions(capabilities) {
                     completed: { type: "boolean", description: "Whether the run is outside an active lifecycle state." },
                     displayOutcome: {
                         anyOf: [
-                            { type: "string", enum: ["VERIFIED", "STOPPED", "NEEDS REVIEW"] },
+                            { type: "string", enum: [...VERIFIED_HANDOFF_OUTCOMES] },
                             { type: "null" }
                         ],
                         description: "Terminal verified-handoff outcome, or null while the run is active."

package/dist/package-version.d.ts

@@ -1 +1 @@
-export declare const MARTIN_MCP_PACKAGE_VERSION = "0.8.4";
+export declare const MARTIN_MCP_PACKAGE_VERSION = "0.8.6";

package/dist/package-version.js

@@ -1,3 +1,3 @@
 // Auto-generated by build — do not edit by hand.
 // Source of truth: packages/mcp/package.json#version
-export const MARTIN_MCP_PACKAGE_VERSION = "0.8.4";
+export const MARTIN_MCP_PACKAGE_VERSION = "0.8.6";

package/dist/server-validation.js

@@ -435,7 +435,7 @@ function validateGetVerificationResultsInput(args) {
 function validateRunDossierInput(args) {
     const record = requireObject(args);
     assertAllowedKeys(record, ["file", "loopId", "runsDir", "latest", "format"]);
-    const base = validateGetRunInput(args);
+    const base = validateGetRunInput(pickKeys(record, RUN_SELECTOR_KEYS));
     return {
         ...base,
         ...optionalEnumAsObject(record.format, "format", ["json", "md", "github-pr"])
@@ -447,7 +447,7 @@ function validateEvalInput(args) {
 function validateRunControlInput(args) {
     const record = requireObject(args);
     assertAllowedKeys(record, ["file", "loopId", "runsDir", "latest", "reason", "requestedBy"]);
-    const base = validateGetRunInput(args);
+    const base = validateGetRunInput(pickKeys(record, RUN_SELECTOR_KEYS));
     return {
         ...base,
         ...optionalString(record.reason, "reason"),
@@ -457,7 +457,7 @@ function validateRunControlInput(args) {
 function validateCreatePrInput(args) {
     const record = requireObject(args);
     assertAllowedKeys(record, ["file", "loopId", "runsDir", "latest", "format", "title", "base", "execute"]);
-    const base = validateRunDossierInput(args);
+    const base = validateRunDossierInput(pickKeys(record, [...RUN_SELECTOR_KEYS, "format"]));
     return {
         ...base,
         ...optionalString(record.title, "title"),
@@ -468,12 +468,17 @@ function validateCreatePrInput(args) {
 function validateReviewPrInput(args) {
     const record = requireObject(args);
     assertAllowedKeys(record, ["file", "loopId", "runsDir", "latest", "format", "prBody"]);
-    const base = validateRunDossierInput(args);
+    const base = validateRunDossierInput(pickKeys(record, [...RUN_SELECTOR_KEYS, "format"]));
     return {
         ...base,
         ...optionalString(record.prBody, "prBody")
     };
 }
+const RUN_SELECTOR_KEYS = ["file", "loopId", "runsDir", "latest"];
+// Base validators check their own allowed keys, so composed validators hand them only those keys.
+function pickKeys(record, keys) {
+    return Object.fromEntries(Object.entries(record).filter(([key]) => keys.includes(key)));
+}
 function requireObject(value) {
     if (!value || typeof value !== "object" || Array.isArray(value)) {
         throw invalidArgumentsError("Tool arguments must be an object.");

package/dist/server.js

@@ -501,6 +501,25 @@ const preflightOutputSchema = {
                     },
                     required: ["available", "detail"]
                 },
+                // Present when the Codex launch probe ran on a live host.
+                codexDiagnostics: {
+                    type: "object",
+                    additionalProperties: true,
+                    properties: {
+                        selectedPath: { type: "string" },
+                        hostPlatform: { type: "string" },
+                        installKind: { type: "string" },
+                        nativeInstallValid: { type: "boolean" },
+                        invocationMode: { type: "string" },
+                        sandboxMode: { type: "string" },
+                        sandboxCompatible: { type: "boolean" },
+                        launchReady: { type: "boolean" },
+                        summary: { type: "string" },
+                        remediation: { type: "string" },
+                        candidateProbeResults: { type: "array", items: { type: "object", additionalProperties: true } }
+                    },
+                    required: ["launchReady", "summary"]
+                },
                 runsRoot: { type: "string" },
                 pathScope: {
                     type: "object",
@@ -775,9 +794,21 @@ const controlOutputSchema = {
         controlId: { type: "string", description: "Unique receipt ID for this control action." },
         requestedAt: { type: "string", description: "ISO 8601 timestamp when the control was recorded." },
         requestedBy: { type: "string", description: "Identity that requested the control, if provided." },
-        reason: { type: "string", description: "Human-readable reason for the control, if provided." }
+        reason: { type: "string", description: "Human-readable reason for the control, if provided." },
+        ok: { type: "boolean", description: "Whether the control receipt was recorded." },
+        summary: { type: "string", description: "Human-readable result of the control request." },
+        requestedAction: {
+            type: "string",
+            enum: ["pause", "cancel", "continue"],
+            description: "The control action that was requested."
+        },
+        state: {
+            type: "object",
+            additionalProperties: true,
+            description: "Run control state after this request, including all recorded receipts."
+        }
     },
-    required: ["loopId", "action", "controlId", "requestedAt"]
+    required: ["loopId", "action", "controlId", "requestedAt", "ok", "summary", "requestedAction", "state"]
 };
 const evalOutputSchema = {
     type: "object",
@@ -830,6 +861,17 @@ const prSummaryOutputSchema = {
     },
     required: ["loopId", "title", "body", "grade", "score"]
 };
+const createPrOutputSchema = {
+    ...prSummaryOutputSchema,
+    properties: {
+        ...prSummaryOutputSchema.properties,
+        execute: { type: "boolean", description: "Whether gh was asked to create the pull request." },
+        created: { type: "boolean", description: "Whether a pull request was created." },
+        branch: { type: "string", description: "Current branch of the run's repository, when known." },
+        url: { type: "string", description: "URL of the created pull request." }
+    },
+    required: [...prSummaryOutputSchema.required, "execute", "created"]
+};
 const prReviewOutputSchema = {
     type: "object",
     additionalProperties: false,
@@ -1505,7 +1547,7 @@ export function createMartinMcpServer(serverInfo) {
                         execute: { type: "boolean" }
                     }
                 },
-                outputSchema: prSummaryOutputSchema
+                outputSchema: createPrOutputSchema
             },
             {
                 name: "martin_review_pr",

package/dist/tools/pr-tools.js

@@ -33,6 +33,18 @@ export async function martinCreatePrTool(input) {
             ...(branch ? { branch } : {})
         };
     }
+    // Opening a PR claims the work is done. Only a VERIFIED run with intact receipts qualifies.
+    const dossier = await martinRunDossierTool(input);
+    const outcome = dossier.verifiedHandoff.outcome;
+    const integrity = dossier.receiptIntegrity.state;
+    if (outcome !== "VERIFIED" || integrity !== "verified") {
+        throw new MartinToolError("policy_blocked", `Refusing to open a PR: run ${summary.loopId} is ${outcome} with receipt integrity ${integrity}.`, {
+            category: "policy_blocked",
+            suggestion: "Open a PR only from a VERIFIED run, or rerun martin_create_pr with execute=false to preview it.",
+            retryable: false,
+            details: { outcome, receiptIntegrity: integrity }
+        });
+    }
     const gh = detectCliAvailability("gh");
     if (!gh.available) {
         throw new MartinToolError("engine_unavailable", "GitHub CLI is not available on PATH.", {

package/dist/tools/run-controls.d.ts

@@ -29,6 +29,11 @@ export interface MartinRunControlOutput {
     loopId: string;
     requestedAction: MartinControlAction;
     state: MartinRunControlState;
+    action: MartinControlAction;
+    controlId: string;
+    requestedAt: string;
+    reason?: string;
+    requestedBy?: string;
 }
 export declare function createRunControlReceipt(action: MartinControlAction, input: MartinRunControlRequestInput): Promise<MartinRunControlOutput>;
 export declare function readRunControlState(detailOrInput: DetailedLoopSource | MartinRunControlRequestInput): Promise<MartinRunControlState>;

package/dist/tools/run-controls.js

@@ -27,7 +27,12 @@ export async function createRunControlReceipt(action, input) {
                 : `Continue request recorded for ${detail.loop.loopId}.`,
         loopId: detail.loop.loopId,
         requestedAction: action,
-        state
+        state,
+        action: receipt.action,
+        controlId: receipt.controlId,
+        requestedAt: receipt.requestedAt,
+        ...(receipt.reason ? { reason: receipt.reason } : {}),
+        ...(receipt.requestedBy ? { requestedBy: receipt.requestedBy } : {})
     };
 }
 export async function readRunControlState(detailOrInput) {

package/dist/vendor/adapters/claude-cli.js

@@ -210,6 +210,39 @@ function extractUsage(parsed, modelLabel, modelSource) {
         }
     });
 }
+/**
+ * Returns the provider's failure message when the Codex JSONL stream shows a failed turn:
+ * an explicit `turn.failed`, or an `error` event with no `turn.completed`. Undefined otherwise.
+ */
+function findCodexTurnFailure(stdout) {
+    let completed = false;
+    let turnFailed = false;
+    let message;
+    for (const line of stdout.split(/\r?\n/u)) {
+        let event;
+        try {
+            event = JSON.parse(line.trim());
+        }
+        catch {
+            continue;
+        }
+        if (event.type === "turn.completed")
+            completed = true;
+        if (event.type === "turn.failed") {
+            turnFailed = true;
+            message = String(event.error?.message ?? message ?? "Codex turn failed.");
+        }
+        else if (event.type === "error") {
+            message ??= String(event.message ?? "Codex reported an error.");
+        }
+    }
+    if (turnFailed)
+        return message;
+    return message !== undefined && !completed ? message : undefined;
+}
+function codexTurnCompleted(stdout) {
+    return /"type"\s*:\s*"turn\.completed"/u.test(stdout);
+}
 function extractCodexJsonlResult(stdout, modelLabel) {
     const events = stdout
         .split(/\r?\n/u)
@@ -829,8 +862,27 @@ export function createAgentCliAdapter(options) {
             const geminiJsonResult = !supportsJsonOutput && options.command === "gemini"
                 ? extractGeminiJsonResult(agentResult.stdout, options.model)
                 : undefined;
+            // A failed Codex turn did no verified work: never let the verifier run against an
+            // untouched workspace and certify it.
+            const codexTurnFailure = options.command === "codex" ? findCodexTurnFailure(agentResult.stdout) : undefined;
+            if (codexTurnFailure !== undefined) {
+                const failureMessage = `codex turn failed: ${codexTurnFailure}`;
+                return {
+                    status: "failed",
+                    summary: "codex turn failed before verifier execution.",
+                    usage: codexJsonlResult?.usage ?? normalizeUsage({
+                        actualUsd: 0,
+                        tokensIn: 0,
+                        tokensOut: 0,
+                        provenance: "unavailable"
+                    }),
+                    verification: { passed: false, summary: `Verifier not run: ${failureMessage}` },
+                    failure: { message: failureMessage }
+                };
+            }
             const producedStructuredCompletion = parsed?.result !== undefined ||
-                codexJsonlResult !== undefined ||
+                (codexJsonlResult !== undefined &&
+                    (agentResult.exitCode === 0 || codexTurnCompleted(agentResult.stdout))) ||
                 geminiJsonResult !== undefined;
             if (agentResult.exitCode !== 0 && !producedStructuredCompletion) {
                 const failureMessage = formatPreVerifierSubprocessFailure(options.command, agentResult.stderr || agentResult.stdout, agentResult.exitCode);

package/dist/vendor/adapters/process-supervisor.d.ts

@@ -57,4 +57,11 @@ export declare function spawnSupervisedProcess(command: string, args: readonly s
 export declare function spawnSupervisedProcessForTest(command: string, args: readonly string[], options: SpawnSupervisedProcessOptions & {
     spawnImpl: ProcessSpawnLike;
 }, seams: ProcessSupervisorTestSeams): Promise<SupervisedProcessResult>;
+/**
+ * Classify an OS process-table snapshot with PID, PGID and STAT columns.
+ * POSIX Z means terminated but not yet reaped; other states remain live.
+ * Malformed or empty output must never assert cleanup success.
+ * @internal - deterministic pure classification for process-supervisor tests
+ */
+export declare function classifyPosixProcessGroupSnapshot(output: string, ownedGroupPid: number): "closed" | "alive" | "absent" | "invalid";
 export {};

package/dist/vendor/adapters/process-supervisor.js

@@ -327,14 +327,21 @@ async function terminatePosixGroupAfterRootExit(pid, killGroup, graceMs, waitImp
     const initial = probePosixProcessGroup(target, killGroup);
     if (initial === "closed")
         return { state: "closed" };
-    if (initial instanceof Error)
+    if (initial instanceof Error) {
+        if (isErrno(initial, "EPERM") && await confirmInactivePosixGroup(pid, killGroup)) {
+            return { state: "closed" };
+        }
         return cleanupFailure("termination_failed", initial);
+    }
     try {
         killGroup(target, "SIGTERM");
     }
     catch (error) {
         if (isErrno(error, "ESRCH"))
             return { state: "closed" };
+        if (isErrno(error, "EPERM") && await confirmInactivePosixGroup(pid, killGroup)) {
+            return { state: "closed" };
+        }
         return cleanupFailure("termination_failed", error);
     }
     if (await waitForPosixGroupExit(target, killGroup, graceMs, waitImpl))
@@ -345,14 +352,30 @@ async function terminatePosixGroupAfterRootExit(pid, killGroup, graceMs, waitImp
     catch (error) {
         if (isErrno(error, "ESRCH"))
             return { state: "closed" };
+        if (isErrno(error, "EPERM") && await confirmInactivePosixGroup(pid, killGroup)) {
+            return { state: "closed" };
+        }
         return cleanupFailure("termination_failed", error);
     }
     if (await waitForPosixGroupExit(target, killGroup, graceMs, waitImpl))
         return { state: "closed" };
+    // A reparented, already-terminated descendant can remain as a zombie in the
+    // process table. On POSIX, kill(-pgid, 0) still sees that process group even
+    // though its remaining members cannot execute code. Prove the absence of
+    // *runnable* group members rather than treating an unreapable zombie as live.
+    // Do not turn an unreadable/ambiguous process snapshot into success.
+    const snapshot = await inspectPosixProcessGroupMembers(pid);
+    if (snapshot === "closed" || (snapshot === "absent" && probePosixProcessGroup(target, killGroup) === "closed")) {
+        return { state: "closed" };
+    }
+    if (snapshot instanceof Error)
+        return cleanupFailure("termination_failed", snapshot);
     return {
         state: "failed",
         code: "tree_close_timeout",
-        message: "Owned POSIX process group remained live after SIGTERM and SIGKILL."
+        message: snapshot === "invalid"
+            ? "Could not prove owned POSIX group closure from the process table."
+            : "Owned POSIX process group still has live members after SIGTERM and SIGKILL."
     };
 }
 async function waitForPosixGroupExit(target, killGroup, durationMs, waitImpl) {
@@ -367,6 +390,14 @@ async function waitForPosixGroupExit(target, killGroup, durationMs, waitImpl) {
     }
     return probePosixProcessGroup(target, killGroup) === "closed";
 }
+async function confirmInactivePosixGroup(ownedGroupPid, killGroup) {
+    const snapshot = await inspectPosixProcessGroupMembers(ownedGroupPid);
+    if (snapshot === "closed")
+        return true; // Observed owned members are all defunct zombies.
+    // An absent group is not proof on its own after EPERM. Require an ESRCH re-probe.
+    return snapshot === "absent"
+        && probePosixProcessGroup(-ownedGroupPid, killGroup) === "closed";
+}
 function probePosixProcessGroup(target, killGroup) {
     try {
         killGroup(target, 0);
@@ -378,6 +409,49 @@ function probePosixProcessGroup(target, killGroup) {
         return error instanceof Error ? error : new Error(String(error));
     }
 }
+/**
+ * Classify an OS process-table snapshot with PID, PGID and STAT columns.
+ * POSIX Z means terminated but not yet reaped; other states remain live.
+ * Malformed or empty output must never assert cleanup success.
+ * @internal - deterministic pure classification for process-supervisor tests
+ */
+export function classifyPosixProcessGroupSnapshot(output, ownedGroupPid) {
+    if (!isOwnedPid(ownedGroupPid))
+        return "invalid";
+    let observed = 0;
+    let matchingGroupMembers = 0;
+    for (const line of output.split(/\r?\n/u)) {
+        if (!line.trim())
+            continue;
+        const match = /^\s*(\d+)\s+(\d+)\s+(\S+)\s*$/u.exec(line);
+        if (!match)
+            return "invalid";
+        observed += 1;
+        if (Number(match[2]) === ownedGroupPid) {
+            matchingGroupMembers += 1;
+            if (!match[3].startsWith("Z"))
+                return "alive";
+        }
+    }
+    if (observed === 0)
+        return "invalid";
+    return matchingGroupMembers > 0 ? "closed" : "absent";
+}
+async function inspectPosixProcessGroupMembers(ownedGroupPid) {
+    return new Promise((resolveResult) => {
+        execFile("ps", ["-axo", "pid=,pgid=,stat="], {
+            encoding: "utf8",
+            timeout: 2_000,
+            maxBuffer: 8 * 1024 * 1024
+        }, (error, stdout) => {
+            if (error) {
+                resolveResult(error);
+                return;
+            }
+            resolveResult(classifyPosixProcessGroupSnapshot(stdout, ownedGroupPid));
+        });
+    });
+}
 function isErrno(error, code) {
     return typeof error === "object" && error !== null && "code" in error
         && String(error.code) === code;

package/dist/vendor/core/grounding.js

@@ -5,6 +5,21 @@ import { homedir } from "node:os";
 import { extname, join, relative } from "node:path";
 import { promisify } from "node:util";
 const execFileAsync = promisify(execFile);
+// Callables provided by the JavaScript language, Node.js, browsers, and common test runners.
+// They are never declared in the repository, so they must not count as ungrounded symbols.
+const STANDARD_CALLABLE_GLOBALS = new Set([
+    // CommonJS / Node.js
+    "require", "setTimeout", "setInterval", "setImmediate", "clearTimeout", "clearInterval",
+    "clearImmediate", "queueMicrotask", "structuredClone", "fetch",
+    // ECMAScript globals
+    "parseInt", "parseFloat", "isNaN", "isFinite", "encodeURIComponent", "decodeURIComponent",
+    "encodeURI", "decodeURI", "escape", "unescape", "eval",
+    // Browser globals
+    "alert", "confirm", "prompt", "atob", "btoa", "requestAnimationFrame", "cancelAnimationFrame",
+    // Test runners (node:test, vitest, jest, mocha, chai)
+    "describe", "suite", "test", "it", "expect", "assert", "before", "after", "beforeEach",
+    "afterEach", "beforeAll", "afterAll", "context", "specify", "should"
+]);
 const TEXT_EXTENSIONS = new Set([
     ".ts", ".tsx", ".js", ".jsx", ".mjs", ".cjs",
     ".json", ".md", ".txt", ".yaml", ".yml", ".py", ".go",
@@ -284,17 +299,30 @@ export function scanPatchForGroundingViolations(diff, index, options = {}) {
         "while",
         "switch",
         "catch",
-        "typeof"
+        "typeof",
+        "super",
+        "yield",
+        "delete",
+        "throw",
+        "instanceof",
+        "constructor"
     ]);
     const addedLines = [...diff.matchAll(addedLinePattern)].map((match) => match[1] ?? "");
     const declaredSymbols = new Set();
     const declarationPattern = /\b(?:function|class|interface|type|enum|const|let|var)\s+([A-Za-z_][A-Za-z0-9_]*)/g;
+    // Class and object-literal method definitions: `name(args) {`, optionally async/static/get/set.
+    const methodDefinitionPattern = /(?:^|[{,])\s*(?:(?:public|private|protected|static|async|get|set|override)\s+)*([A-Za-z_][A-Za-z0-9_]*)\s*\([^()]*\)\s*(?::[^{=]+)?\{/gu;
     for (const line of addedLines) {
         for (const declarationMatch of line.matchAll(declarationPattern)) {
             const symbol = declarationMatch[1];
             if (symbol)
                 declaredSymbols.add(symbol.toLowerCase());
         }
+        for (const methodMatch of line.matchAll(methodDefinitionPattern)) {
+            const methodName = methodMatch[1];
+            if (methodName && !symbolKeywords.has(methodName))
+                declaredSymbols.add(methodName.toLowerCase());
+        }
         if (/^\s*import\b/u.test(line)) {
             const importClause = line.match(/^\s*import\s+(.+?)\s+from\s+['"]/u)?.[1] ?? "";
             for (const importSymbol of importClause.matchAll(/\b(?:as\s+)?([A-Za-z_][A-Za-z0-9_]*)\b/g)) {
@@ -334,6 +362,8 @@ export function scanPatchForGroundingViolations(diff, index, options = {}) {
                 continue;
             if (symbolKeywords.has(symbol))
                 continue;
+            if (STANDARD_CALLABLE_GLOBALS.has(symbol))
+                continue;
             if (/^[A-Z]/u.test(symbol))
                 continue;
             if (declaredSymbols.has(symbol.toLowerCase()))

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@martinloop/mcp",
-  "version": "0.8.4",
+  "version": "0.8.6",
   "mcpName": "io.github.Keesan12/martin-loop",
   "private": false,
   "type": "module",

package/server.json

@@ -7,12 +7,12 @@
     "url": "https://github.com/Keesan12/martin-loop",
     "source": "github"
   },
-  "version": "0.8.4",
+  "version": "0.8.6",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@martinloop/mcp",
-      "version": "0.8.4",
+      "version": "0.8.6",
       "transport": {
         "type": "stdio"
       }