@martinloop/mcp 0.3.9 → 0.5.1

package/dist/tools/run-loop.js

@@ -1,4 +1,4 @@
-import { createClaudeCliAdapter, createCodexCliAdapter, createGeminiCliAdapter, probeCodexLaunch, resolveCliCommandAvailability, createVerifierOnlyAdapter } from "../vendor/adapters/index.js";
+import { createClaudeCliAdapter, createCodexCliAdapter, createGeminiCliAdapter, createVerifierOnlyAdapter, probeCodexLaunch, resolveCliCommandAvailability } from "../vendor/adapters/index.js";
 import { createFileRunStore, evaluateCostGovernor, resolveRunsRoot, runMartin } from "../vendor/core/index.js";
 import { normalizeSafePathPatterns, resolveSafeRepoRoot } from "../server-validation.js";
 import { MartinToolError } from "./tool-errors.js";
@@ -6,16 +6,12 @@ import { buildArtifactSummary, buildVerificationSummary, buildLoopPreview, build
 import { normalizeLoopBudget } from "./workflow-governance.js";
 let proofModeVerifierSpawnImpl;
 let runStoreOverrideForTests;
-let runMartinImpl = runMartin;
 export function __setProofModeVerifierSpawnImplForTests(spawnImpl) {
     proofModeVerifierSpawnImpl = spawnImpl;
 }
 export function __setRunStoreOverrideForTests(store) {
     runStoreOverrideForTests = store;
 }
-export function __setRunMartinImplForTests(impl) {
-    runMartinImpl = impl ?? runMartin;
-}
 export async function runLoopTool(input) {
     const workingDirectory = resolveSafeRepoRoot(input.workingDirectory);
     const engine = input.engine ?? "claude";
@@ -23,6 +19,9 @@ export async function runLoopTool(input) {
     const allowedPaths = normalizeSafePathPatterns(input.allowedPaths, "allowedPaths");
     const deniedPaths = normalizeSafePathPatterns(input.deniedPaths, "deniedPaths");
     const executionMode = resolveExecutionMode();
+    const evidenceExecutionMode = executionMode.liveMode
+        ? "governed"
+        : "verification_only";
     const workspaceRoot = resolveSafeRepoRoot();
     const runsRoot = resolveRunsRoot(process.env);
     const receiptScope = {
@@ -38,7 +37,7 @@ export async function runLoopTool(input) {
             if (!engineAvailability.available) {
                 throw new MartinToolError("engine_unavailable", `Engine '${engine}' is not available on PATH.`, {
                     category: "environment",
-                    suggestion: "Install the requested CLI or set MARTIN_LIVE=false for a no-spend proof run.",
+                    suggestion: "Install the requested CLI. MARTIN_LIVE=false is verification-only and cannot emit governed VERIFIED.",
                     retryable: false
                 });
             }
@@ -60,7 +59,7 @@ export async function runLoopTool(input) {
             if (!engineAvailability.available) {
                 throw new MartinToolError("engine_unavailable", `Engine '${engine}' is not available on PATH.`, {
                     category: "environment",
-                    suggestion: "Install the requested CLI or set MARTIN_LIVE=false for a no-spend proof run.",
+                    suggestion: "Install the requested CLI. MARTIN_LIVE=false is verification-only and cannot emit governed VERIFIED.",
                     retryable: false
                 });
             }
@@ -69,7 +68,7 @@ export async function runLoopTool(input) {
     const adapter = !executionMode.liveMode
         ? createVerifierOnlyAdapter({
             workingDirectory,
-            label: "Proof mode adapter (MARTIN_LIVE=false)",
+            label: "Verification-only adapter (MARTIN_LIVE=false)",
             ...(input.verifyTimeoutMs !== undefined ? { verifyTimeoutMs: input.verifyTimeoutMs } : {}),
             ...(proofModeVerifierSpawnImpl ? { spawnImpl: proofModeVerifierSpawnImpl } : {})
         })
@@ -102,7 +101,7 @@ export async function runLoopTool(input) {
         partialBudget.maxTokens = input.maxTokens;
     }
     const budget = normalizeLoopBudget(partialBudget);
-    const result = await runMartinImpl({
+    const result = await runMartin({
         workspaceId: input.workspaceId ?? "ws_mcp",
         projectId: input.projectId ?? "proj_mcp",
         store: runStoreOverrideForTests ?? createFileRunStore({ runsRoot }),
@@ -119,6 +118,10 @@ export async function runLoopTool(input) {
             ...(deniedPaths ? { deniedPaths } : {})
         },
         budget,
+        metadata: {
+            executionMode: evidenceExecutionMode,
+            governanceClaimEligible: evidenceExecutionMode === "governed" ? "true" : "false"
+        },
         adapter
     });
     const lastAttempt = result.loop.attempts.at(-1);
@@ -142,7 +145,10 @@ export async function runLoopTool(input) {
         reason: result.decision.reason,
         attempts: result.loop.attempts.length,
         costUsd: result.loop.cost.actualUsd,
+        costProvenance: result.loop.cost.provenance ?? "unavailable",
         verificationPassed,
+        executionMode: evidenceExecutionMode,
+        governanceClaimEligible: evidenceExecutionMode === "governed",
         loopId: result.loop.loopId,
         pressure: costState.pressure,
         shouldStop: costState.shouldStop,

package/dist/tools/tool-errors.js

@@ -92,7 +92,7 @@ export function toToolFailure(error) {
             code: "engine_unavailable",
             category: "environment",
             message,
-            suggestion: "Install the requested CLI or set MARTIN_LIVE=false for a no-spend proof run.",
+            suggestion: "Install the requested CLI. MARTIN_LIVE=false is verification-only and cannot emit governed VERIFIED.",
             retryable: false
         };
     }

package/dist/tools/tool-response.js

@@ -4,13 +4,20 @@ function createJsonBlock(output) {
         text: JSON.stringify(output, null, 2)
     };
 }
+function resolveHumanArtifact(output, summary) {
+    const rendered = "rendered" in output ? output.rendered : undefined;
+    return typeof rendered === "string" && rendered.trim().length > 0 ? rendered : summary;
+}
 export function createToolSuccessResult(output, summary) {
     return {
-        content: [createJsonBlock(output), { type: "text", text: summary }],
+        content: [
+            { type: "text", text: resolveHumanArtifact(output, summary) },
+            createJsonBlock(output)
+        ],
         structuredContent: output,
         _meta: {
             "martinloop/summary": summary,
-            "martinloop/contentVersion": "2026-05-15"
+            "martinloop/contentVersion": "2026-08-18"
         }
     };
 }

package/dist/tools/tool-support.d.ts

@@ -1,4 +1,4 @@
-import type { LoopArtifact, LoopBudget, LoopCost, LoopEvent, LoopTask, ReceiptIntegritySummary, ReceiptScope } from "../vendor/contracts/index.js";
+import type { CostProvenance, LoopArtifact, LoopBudget, LoopCost, LoopEvent, LoopTask, ReceiptIntegritySummary, ReceiptScope, TerminationEnvelopeV1 } from "../vendor/contracts/index.js";
 import { type LedgerEvent, type LoopAttemptRecord, type LoopRunRecord } from "../vendor/core/index.js";
 export declare const MARTIN_ENGINE_VALUES: readonly ["claude", "codex", "gemini", "openai"];
 export type MartinEngine = (typeof MARTIN_ENGINE_VALUES)[number];
@@ -14,6 +14,7 @@ export interface InspectableLoopRecord extends Omit<LoopRunRecord, "attempts" |
     metadata?: Record<string, string>;
     receiptIntegrity?: ReceiptIntegritySummary;
     receiptScope?: ReceiptScope;
+    terminationEnvelope?: TerminationEnvelopeV1;
     routingEconomics?: import("../vendor/contracts/index.js").RoutingEconomics;
 }
 export interface LoopPreview {
@@ -26,6 +27,7 @@ export interface LoopPreview {
     updatedAt?: string;
     attempts: number;
     costUsd: number;
+    costProvenance: CostProvenance;
     avoidedUsd: number;
     pressure: string;
     shouldStop: boolean;
@@ -127,7 +129,7 @@ export declare function inspectRunsRoot(runsRoot?: string): Promise<RunStoreInsp
 export declare function buildRunRecordPaths(runsRoot: string, loopId: string): CanonicalRunPaths;
 export declare function buildAttemptArtifactDirectory(runsRoot: string, loopId: string, attemptIndex: number): string;
 export declare function buildAttemptArtifactsReference(runsRoot: string, loopId: string, attemptIndex: number): Promise<AttemptArtifactFiles>;
-export declare function buildCostSnapshot(cost: Pick<LoopCost, "actualUsd" | "tokensIn" | "tokensOut"> & {
+export declare function buildCostSnapshot(cost: Pick<LoopCost, "actualUsd" | "tokensIn" | "tokensOut" | "provenance"> & {
     avoidedUsd?: number;
 }): LoopCost;
 export declare function buildBudgetSnapshot(budget: LoopBudget): LoopBudget;

package/dist/tools/tool-support.js

@@ -15,7 +15,7 @@ export function resolveExecutionMode() {
         mode: liveMode ? "live" : "proof",
         detail: liveMode
             ? "Live CLI execution is enabled."
-            : "Proof mode is active because MARTIN_LIVE=false."
+            : "Verification-only mode is active because MARTIN_LIVE=false; governed VERIFIED is unavailable."
     };
 }
 export function detectCliAvailability(command) {
@@ -105,11 +105,6 @@ function discoverCommonInstallDirectories(command) {
         if (localAppData) {
             dirs.push(join(localAppData, "OpenAI", "Codex", "bin"));
         }
-        // Claude Code native installer places binary at %USERPROFILE%\.local\bin
-        const userProfile = process.env.USERPROFILE ?? process.env.HOMEPATH;
-        if (userProfile) {
-            dirs.push(join(userProfile, ".local", "bin"));
-        }
         // Scoop
         if (home) {
             dirs.push(join(home, "scoop", "shims"));
@@ -166,13 +161,8 @@ function isOnPathDirectly(command, resolvedPath) {
  * Returns a platform-specific one-liner install command for a known CLI.
  */
 function suggestInstallCommand(command) {
-    if (command === "claude") {
-        const installCmd = process.platform === "win32"
-            ? "irm https://claude.ai/install.ps1 | iex"
-            : "curl -fsSL https://claude.ai/install.sh | bash";
-        return `Install with: ${installCmd}`;
-    }
     const npmInstalls = {
+        claude: "npm install -g @anthropic-ai/claude-code",
         codex: "npm install -g @openai/codex",
         gemini: "npm install -g @google/gemini-cli"
     };
@@ -185,7 +175,7 @@ function suggestInstallCommand(command) {
 export function getEngineAvailability(engine) {
     return detectCliAvailability(engine);
 }
-export function createSkippedCliAvailability(command, detail = "Proof mode skipped live CLI availability detection.") {
+export function createSkippedCliAvailability(command, detail = "Verification-only mode skipped live CLI availability detection; governed VERIFIED is unavailable.") {
     return {
         command,
         available: false,
@@ -218,6 +208,7 @@ export function buildLoopPreview(loop) {
         ...(loop.updatedAt ? { updatedAt: loop.updatedAt } : {}),
         attempts: loop.attempts.length,
         costUsd: loop.cost.actualUsd,
+        costProvenance: loop.cost.provenance ?? "unavailable",
         avoidedUsd: loop.cost.avoidedUsd ?? 0,
         pressure: costState.pressure,
         shouldStop: costState.shouldStop,
@@ -421,7 +412,8 @@ export function buildCostSnapshot(cost) {
         actualUsd: cost.actualUsd,
         avoidedUsd: cost.avoidedUsd ?? 0,
         tokensIn: cost.tokensIn,
-        tokensOut: cost.tokensOut
+        tokensOut: cost.tokensOut,
+        provenance: cost.provenance ?? "unavailable"
     };
 }
 export function buildBudgetSnapshot(budget) {
@@ -439,7 +431,6 @@ export function buildSuggestedResourceUris(loopId) {
         "martin://runs/triage",
         "martin://runs/latest",
         "martin://runs/latest/summary",
-        "martin://runs/latest/receipt",
         "martin://runs/latest/proof-card",
         "martin://runs/latest/budget-status",
         "martin://runs/latest/verifier-evidence",

package/dist/vendor/adapters/claude-cli.d.ts

@@ -96,7 +96,7 @@ export interface GeminiCliAdapterOptions {
     timeoutMs?: number;
     verifyTimeoutMs?: number;
     label?: string;
-    /** Override the model passed via --model flag. Defaults to the Gemini `flash` alias. */
+    /** Explicit model override passed via --model. Omitted to preserve Gemini Auto. */
     model?: string;
     /** Approval mode for headless Gemini runs. Defaults to yolo for autonomous execution. */
     approvalMode?: "default" | "auto_edit" | "yolo" | "plan";
@@ -135,7 +135,7 @@ export declare function createClaudeCliAdapter(options?: ClaudeCliAdapterOptions
  */
 export declare function createCodexCliAdapter(options?: CodexCliAdapterOptions): MartinAdapter;
 /**
- * Spawns `gemini --model <model> --prompt "" --approval-mode <mode> --output-format json [...]`.
+ * Spawns `gemini [--model <explicit-model>] --prompt "" --approval-mode <mode> --output-format json [...]`.
  *
  * The prompt is delivered via stdin while forcing headless mode with `--prompt ""`,
  * which keeps large MartinLoop prompts off the command line on Windows.

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

@@ -11,8 +11,8 @@
  *
  * MCP tools and integration tests use the same factories.
  */
-import { readGitExecutionArtifacts, resolveGitRepositoryRoot, runSubprocess, runVerification } from "./cli-bridge.js";
-import { buildCodexExecArgs, DEFAULT_CODEX_CHATGPT_MODEL } from "./codex-launcher.js";
+import { readGitChangedFiles, readGitExecutionArtifacts, resolveGitRepositoryRoot, runSubprocess, runVerification } from "./cli-bridge.js";
+import { buildCodexExecArgs } from "./codex-launcher.js";
 import { createAdapterCapabilities, normalizeStructuredErrors, normalizeUsage } from "./runtime-support.js";
 // ---------------------------------------------------------------------------
 // Cost estimation
@@ -68,7 +68,22 @@ function calculateUsageCost(usage, pricing) {
         (usage.cacheCreationInputTokens / 1000) * (pricing.cacheCreationInputPer1K ?? pricing.inputPer1K) +
         (usage.outputTokens / 1000) * pricing.outputPer1K);
 }
-function extractUsage(parsed, modelLabel) {
+const EMBEDDED_PRICING_VERSION = "embedded-v1";
+function extractClaudeObservedModel(stdout) {
+    for (const line of stdout.split(/\r?\n/u)) {
+        try {
+            const event = JSON.parse(line);
+            if (event.type === "system" && event.subtype === "init" && typeof event.model === "string") {
+                return event.model;
+            }
+        }
+        catch {
+            // Ignore non-JSON stream fragments.
+        }
+    }
+    return undefined;
+}
+function extractUsage(parsed, modelLabel, modelSource) {
     if (!parsed?.usage) {
         return normalizeUsage({
             actualUsd: 0,
@@ -102,15 +117,24 @@ function extractUsage(parsed, modelLabel) {
         tokensIn,
         tokensOut,
         cachedInputTokens,
-        provenance: hasAuthoritativeCost ? "actual" : pricingResolution.pricing ? "estimated" : "unavailable",
+        provenance: hasAuthoritativeCost ? "actual" : pricingResolution.pricing ? "calculated" : "unavailable",
         providerSettlement: {
             providerId: "claude",
-            model: modelLabel ?? "claude",
+            ...(modelLabel ? { model: modelLabel } : {}),
             transport: "cli",
             source: "claude_json",
             inputTokens: promptTokens + cacheCreationInputTokens,
             cachedInputTokens,
+            cacheCreationInputTokens,
             outputTokens: tokensOut,
+            billingMode: "unknown",
+            modelSource,
+            pricingSource: hasAuthoritativeCost
+                ? "provider_reported_total"
+                : pricingResolution.pricing
+                    ? "static_catalog"
+                    : "none",
+            ...(pricingResolution.pricing ? { pricingVersion: EMBEDDED_PRICING_VERSION } : {}),
             rawUsageAvailable: true,
             settledAt: new Date().toISOString()
         }
@@ -152,7 +176,7 @@ function extractCodexJsonlResult(stdout, modelLabel) {
                 provenance: "unavailable",
                 providerSettlement: {
                     providerId: "codex",
-                    model: modelLabel ?? "codex",
+                    ...(modelLabel ? { model: modelLabel } : {}),
                     transport: "cli",
                     source: "unavailable",
                     inputTokens: 0,
@@ -169,8 +193,8 @@ function extractCodexJsonlResult(stdout, modelLabel) {
     const reasoningOutputTokens = latestTurnCompleted.usage.reasoning_output_tokens ?? 0;
     const tokensIn = promptTokens + cachedInputTokens;
     const tokensOut = outputTokens + reasoningOutputTokens;
-    const pricing = (modelLabel ? MODEL_PRICING[modelLabel] : undefined) ??
-        MODEL_PRICING["codex"] ??
+    const exactPricing = modelLabel ? MODEL_PRICING[modelLabel] : undefined;
+    const pricing = exactPricing ?? MODEL_PRICING["codex"] ??
         { inputPer1K: BLENDED_INPUT_COST_PER_1K, outputPer1K: BLENDED_OUTPUT_COST_PER_1K };
     const actualUsd = (promptTokens / 1000) * pricing.inputPer1K +
         (cachedInputTokens / 1000) * (pricing.cachedInputPer1K ?? pricing.inputPer1K) +
@@ -183,16 +207,20 @@ function extractCodexJsonlResult(stdout, modelLabel) {
             tokensOut,
             cachedInputTokens,
             reasoningTokensOut: reasoningOutputTokens,
-            provenance: "actual",
+            provenance: exactPricing ? "calculated" : "estimated",
             providerSettlement: {
                 providerId: "codex",
-                model: modelLabel ?? "codex",
+                ...(modelLabel ? { model: modelLabel } : {}),
                 transport: "cli",
                 source: "codex_jsonl",
                 inputTokens: promptTokens,
                 cachedInputTokens,
                 outputTokens,
                 reasoningOutputTokens,
+                billingMode: "unknown",
+                modelSource: modelLabel ? "explicit_override" : "agent_default",
+                pricingSource: exactPricing ? "static_catalog" : "blended_fallback",
+                pricingVersion: EMBEDDED_PRICING_VERSION,
                 rawUsageAvailable: true,
                 settledAt: new Date().toISOString()
             }
@@ -228,7 +256,7 @@ function extractGeminiJsonResult(stdout, modelLabel) {
                 provenance: "unavailable",
                 providerSettlement: {
                     providerId: "gemini",
-                    model: modelLabel ?? "flash",
+                    ...(modelLabel ? { model: modelLabel } : {}),
                     transport: "cli",
                     source: "unavailable",
                     inputTokens: 0,
@@ -254,16 +282,20 @@ function extractGeminiJsonResult(stdout, modelLabel) {
             tokensOut,
             cachedInputTokens,
             reasoningTokensOut: reasoningOutputTokens,
-            provenance: "actual",
+            provenance: "estimated",
             providerSettlement: {
                 providerId: "gemini",
-                model: modelLabel ?? "flash",
+                ...(modelLabel ? { model: modelLabel } : {}),
                 transport: "cli",
                 source: "gemini_json",
                 inputTokens: promptTokens,
                 cachedInputTokens,
                 outputTokens,
                 reasoningOutputTokens,
+                billingMode: "unknown",
+                modelSource: modelLabel ? "explicit_override" : "agent_default",
+                pricingSource: "blended_fallback",
+                pricingVersion: EMBEDDED_PRICING_VERSION,
                 rawUsageAvailable: true,
                 settledAt: new Date().toISOString()
             }
@@ -381,14 +413,10 @@ function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate)
         catch {
             return;
         }
-        // result events contain aggregate usage that duplicates previously streamed
-        // assistant-message usage events — skip to avoid double-counting.
+        extractUsageFromEvent(event, terminate);
         if (event.type === "result") {
             finalResult = event;
         }
-        else {
-            extractUsageFromEvent(event, terminate);
-        }
     };
     return {
         onChunk: (chunk, terminate) => {
@@ -501,7 +529,7 @@ export function createAgentCliAdapter(options) {
         label: options.label ?? `${options.command} CLI adapter`,
         metadata: {
             providerId: options.command,
-            model: options.model ?? options.command,
+            ...(options.model ? { model: options.model } : {}),
             transport: "cli",
             capabilities: createAdapterCapabilities({
                 preflight: true,
@@ -514,6 +542,14 @@ export function createAgentCliAdapter(options) {
         async execute(request) {
             const prompt = buildPrompt(request);
             const estimatedUsage = estimateUsage(prompt, options.model ?? options.command, options.command);
+            const repoRoot = request.context.repoRoot;
+            const gitRepoRoot = repoRoot ? resolveGitRepositoryRoot(repoRoot) : undefined;
+            // A governed run may begin in a deliberately dirty workspace. Capture that
+            // baseline so existing operator work is neither reported as this run's
+            // execution nor treated as scope creep.
+            const baselineChangedFiles = gitRepoRoot
+                ? new Set(await readGitChangedFiles(gitRepoRoot, 5_000, options.spawnImpl))
+                : new Set();
             // Preflight: bail if projected cost exceeds remaining budget
             if (request.context.remainingBudgetUsd > 0) {
                 const projected = estimatePromptCost(prompt, options.model ?? "", options.command);
@@ -549,7 +585,8 @@ export function createAgentCliAdapter(options) {
                 timeoutMs,
                 spawnImpl: options.spawnImpl,
                 ...(stdinData === undefined ? {} : { stdinData }),
-                ...(streamingUsage ? { onStdoutChunk: streamingUsage.onChunk } : {})
+                ...(streamingUsage ? { onStdoutChunk: streamingUsage.onChunk } : {}),
+                ...(request.signal !== undefined ? { signal: request.signal } : {})
             });
             if (agentResult.terminationReason) {
                 const snapshot = streamingUsage?.snapshot();
@@ -660,8 +697,15 @@ export function createAgentCliAdapter(options) {
                 parsed?.result ??
                 agentResult.stdout.trim();
             const summary = truncate(agentText, 2000);
+            const observedClaudeModel = options.streamingUsageCap
+                ? extractClaudeObservedModel(agentResult.stdout)
+                : undefined;
             const usage = parsed?.usage
-                ? extractUsage(parsed, options.model)
+                ? extractUsage(parsed, observedClaudeModel ?? options.model, observedClaudeModel
+                    ? "provider_reported"
+                    : options.model
+                        ? "explicit_override"
+                        : "agent_default")
                 : codexJsonlResult?.usage ??
                     geminiJsonResult?.usage ??
                     normalizeUsage({
@@ -673,7 +717,7 @@ export function createAgentCliAdapter(options) {
                         providerSettlement: options.command === "codex"
                             ? {
                                 providerId: "codex",
-                                model: options.model ?? "codex",
+                                ...(options.model ? { model: options.model } : {}),
                                 transport: "cli",
                                 source: "estimated_fallback",
                                 inputTokens: estimatedUsage.tokensIn,
@@ -684,7 +728,7 @@ export function createAgentCliAdapter(options) {
                             : options.command === "gemini"
                                 ? {
                                     providerId: "gemini",
-                                    model: options.model ?? "flash",
+                                    ...(options.model ? { model: options.model } : {}),
                                     transport: "cli",
                                     source: "estimated_fallback",
                                     inputTokens: estimatedUsage.tokensIn,
@@ -695,30 +739,36 @@ export function createAgentCliAdapter(options) {
                                 : undefined
                     });
             const verificationStack = request.context.verificationStack;
-            const verification = await runVerification(request.context.verificationPlan, workingDirectory, verifyTimeoutMs, verificationStack, options.spawnImpl);
+            const verification = await runVerification(request.context.verificationPlan, workingDirectory, verifyTimeoutMs, verificationStack, options.spawnImpl, {
+                runId: request.loopId,
+                workspaceId: request.workspaceId,
+                cwd: workingDirectory,
+            });
             // Check for zero-diff (agent ran but made no file changes)
-            const repoRoot = request.context.repoRoot;
-            const gitRepoRoot = repoRoot ? resolveGitRepositoryRoot(repoRoot) : undefined;
-            let noDiff = false;
-            if (gitRepoRoot) {
-                noDiff = await checkNoDiff(gitRepoRoot, options.spawnImpl);
-            }
+            const postRunChangedFiles = gitRepoRoot
+                ? await readGitChangedFiles(gitRepoRoot, 5_000, options.spawnImpl)
+                : [];
+            const agentChangedFiles = postRunChangedFiles.filter((file) => !baselineChangedFiles.has(file));
+            const noDiff = gitRepoRoot !== undefined && agentChangedFiles.length === 0;
             // Extract structured errors from stderr/stdout for better failure context
             const structuredErrors = normalizeStructuredErrors(extractStructuredErrors(agentResult.stderr, agentResult.stdout));
-            const executionArtifacts = gitRepoRoot
+            const rawExecutionArtifacts = gitRepoRoot
                 ? await readGitExecutionArtifacts(gitRepoRoot, 5000, options.spawnImpl)
                 : undefined;
+            const executionArtifacts = rawExecutionArtifacts
+                ? {
+                    ...(agentChangedFiles.length > 0 ? { changedFiles: agentChangedFiles } : {}),
+                    ...(baselineChangedFiles.size === 0 && rawExecutionArtifacts.diffStats
+                        ? { diffStats: rawExecutionArtifacts.diffStats }
+                        : {})
+                }
+                : undefined;
             // Scope contract enforcement: check touched files against allowedPaths/deniedPaths
             let scopeViolations = [];
             const scopeCtx = request.context;
             if (gitRepoRoot && (scopeCtx.allowedPaths?.length || scopeCtx.deniedPaths?.length)) {
-                const diffResult = await runSubprocess("git", ["diff", "--name-only", "HEAD"], {
-                    cwd: gitRepoRoot,
-                    timeoutMs: 5000,
-                    spawnImpl: options.spawnImpl
-                });
-                if (diffResult.exitCode === 0 && diffResult.stdout.trim()) {
-                    const touchedFiles = diffResult.stdout.trim().split("\n").filter(Boolean);
+                if (agentChangedFiles.length > 0) {
+                    const touchedFiles = agentChangedFiles;
                     const allowed = scopeCtx.allowedPaths ?? [];
                     const denied = scopeCtx.deniedPaths ?? [];
                     for (const file of touchedFiles) {
@@ -741,7 +791,7 @@ export function createAgentCliAdapter(options) {
                     status: "completed",
                     summary,
                     usage,
-                    verification: { passed: true, summary: verification.summary },
+                    verification,
                     ...(executionArtifacts
                         ? {
                             execution: {
@@ -786,7 +836,7 @@ export function createAgentCliAdapter(options) {
                 }
                 // Reset tracked files to HEAD so next attempt starts from clean state
                 try {
-                    if (gitRepoRoot) {
+                    if (gitRepoRoot && baselineChangedFiles.size === 0) {
                         await runSubprocess("git", ["restore", "--staged", "--worktree", "."], {
                             cwd: gitRepoRoot,
                             timeoutMs: 5000
@@ -803,7 +853,7 @@ export function createAgentCliAdapter(options) {
                     ? `${summary}${errorBlock}${scopeViolations.length > 0 ? `\nScope violations: ${scopeViolations.join(", ")}` : ""}`
                     : summary,
                 usage,
-                verification: { passed: false, summary: verification.summary },
+                verification,
                 ...(executionArtifacts
                     ? {
                         execution: {
@@ -819,18 +869,6 @@ export function createAgentCliAdapter(options) {
                     ...(classHint ? { classHint } : {})
                 }
             };
-        },
-        /**
-         * Return a new adapter instance with a different model.
-         * Used by run-martin.ts when a change_model intervention fires.
-         * Model escalation order: haiku → sonnet → opus (cheapest-first, escalate on repeated failure).
-         */
-        withModel(newModel) {
-            return createAgentCliAdapter({
-                ...options,
-                model: newModel,
-                adapterIdSuffix: `${options.adapterIdSuffix ?? options.command}:${newModel}`
-            });
         }
     };
     return adapter;
@@ -859,7 +897,7 @@ export function createClaudeCliAdapter(options = {}) {
     return createAgentCliAdapter({
         command: "claude",
         adapterIdSuffix: "claude",
-        model: options.model ?? "claude-sonnet-4-6",
+        model: options.model,
         label: options.label ?? "Claude CLI adapter",
         workingDirectory: options.workingDirectory,
         timeoutMs: options.timeoutMs,
@@ -906,11 +944,11 @@ export function createCodexCliAdapter(options = {}) {
     const sandbox = options.sandbox ?? "workspace-write";
     const workingDirectory = options.workingDirectory ?? process.cwd();
     const command = options.command ?? "codex";
-    const launchModel = options.model ?? DEFAULT_CODEX_CHATGPT_MODEL;
+    const launchModel = options.model;
     return createAgentCliAdapter({
         command,
         adapterIdSuffix: "codex",
-        model: options.model ?? "codex",
+        model: options.model,
         label: options.label ?? "Codex CLI adapter",
         workingDirectory,
         timeoutMs: options.timeoutMs,
@@ -931,7 +969,7 @@ export function createCodexCliAdapter(options = {}) {
 // Pre-configured: Gemini CLI
 // ---------------------------------------------------------------------------
 /**
- * Spawns `gemini --model <model> --prompt "" --approval-mode <mode> --output-format json [...]`.
+ * Spawns `gemini [--model <explicit-model>] --prompt "" --approval-mode <mode> --output-format json [...]`.
  *
  * The prompt is delivered via stdin while forcing headless mode with `--prompt ""`,
  * which keeps large MartinLoop prompts off the command line on Windows.
@@ -940,13 +978,12 @@ export function createCodexCliAdapter(options = {}) {
  *   npm install -g @google/gemini-cli
  */
 export function createGeminiCliAdapter(options = {}) {
-    const model = options.model ?? "flash";
     const approvalMode = options.approvalMode ?? "yolo";
     const extraArgs = options.extraArgs ?? [];
     return createAgentCliAdapter({
         command: "gemini",
         adapterIdSuffix: "gemini",
-        model,
+        model: options.model,
         label: options.label ?? "Gemini CLI adapter",
         workingDirectory: options.workingDirectory,
         timeoutMs: options.timeoutMs,
@@ -954,8 +991,7 @@ export function createGeminiCliAdapter(options = {}) {
         supportsJsonOutput: false,
         spawnImpl: options.spawnImpl,
         argsBuilder: () => [
-            "--model",
-            model,
+            ...(options.model ? ["--model", options.model] : []),
             "--prompt",
             "",
             "--approval-mode",
@@ -1152,11 +1188,3 @@ function extractStructuredErrors(stderr, stdout) {
     }
     return errors.slice(0, 10); // cap at 10 to avoid bloating prompts
 }
-async function checkNoDiff(repoRoot, spawnImpl) {
-    const result = await runSubprocess("git", ["diff", "--name-only", "HEAD"], {
-        cwd: repoRoot,
-        timeoutMs: 5000,
-        spawnImpl
-    });
-    return result.exitCode === 0 && result.stdout.trim().length === 0;
-}