ccqa 0.3.7 → 0.3.8

package/README.md

@@ -267,6 +267,8 @@ ccqa generate-setup <name>         Generate and validate setup test script
   --auto / --no-interactive         Same semantics as `generate`
 ```
 
+All Claude-driven commands (`trace`, `trace-setup`, `generate`, `generate-setup`) accept `-m, --model <name>` to select the Claude model — pass an alias (`sonnet` | `opus` | `haiku`) or a full model ID (e.g. `claude-opus-4-7`). The flag overrides the `CCQA_MODEL` environment variable; when both are unset, the Claude Code CLI default is used. Authentication is handled by your local Claude Code login — no `ANTHROPIC_API_KEY` is required.
+
 `<feature/spec>` is a 2-segment alias for the on-disk path `.ccqa/features/<feature>/test-cases/<spec>/`. Pass the alias, not the full directory path.
 
 ## File structure

package/dist/bin/ccqa.mjs

@@ -339,8 +339,14 @@ async function timedPhase(label, fn, scope = "fix") {
 }
 //#endregion
 //#region src/claude/invoke.ts
+function resolveModel(explicit) {
+	if (explicit) return explicit;
+	const envModel = process.env["CCQA_MODEL"];
+	return envModel && envModel.length > 0 ? envModel : void 0;
+}
 async function invokeClaudeStreaming(options, onEvent) {
-	const { prompt, systemPrompt, allowedTools, disableBuiltinTools = false, maxTurns, env, onAbAction, onAbActionFailed } = options;
+	const { prompt, systemPrompt, allowedTools, disableBuiltinTools = false, maxTurns, env, model, onAbAction, onAbActionFailed } = options;
+	const resolvedModel = resolveModel(model);
 	let lastAbToolUseId = null;
 	const sdkOptions = {
 		systemPrompt,
@@ -348,6 +354,7 @@ async function invokeClaudeStreaming(options, onEvent) {
 		allowedTools: allowedTools ?? ["Bash(*)"],
 		permissionMode: "bypassPermissions",
 		allowDangerouslySkipPermissions: true,
+		...resolvedModel ? { model: resolvedModel } : {},
 		...env ? { env: {
 			...process.env,
 			...env
@@ -908,11 +915,11 @@ function envRefsToJsExpression(value) {
 }
 //#endregion
 //#region src/cli/trace.ts
-const traceCommand = new Command("trace").argument("<feature/spec>", "Spec id in '<feature>/<spec>' form (resolves to .ccqa/features/<feature>/test-cases/<spec>/)").description("Run agent-browser, verify assertions, and record structured actions").action(async (specPath) => {
+const traceCommand = new Command("trace").argument("<feature/spec>", "Spec id in '<feature>/<spec>' form (resolves to .ccqa/features/<feature>/test-cases/<spec>/)").description("Run agent-browser, verify assertions, and record structured actions").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").action(async (specPath, opts) => {
 	const { featureName, specName } = parseSpecPath(specPath);
-	await runTrace(featureName, specName);
+	await runTrace(featureName, specName, opts.model);
 });
-async function runTrace(featureName, specName) {
+async function runTrace(featureName, specName, model) {
 	header("trace", `${featureName}/${specName}`);
 	await ensureCcqaDir();
 	const spec = parseTestSpec(await readSpecFile(featureName, specName));
@@ -951,6 +958,7 @@ async function runTrace(featureName, specName) {
 			AGENT_BROWSER_SESSION: sessionName,
 			PATH: pathWithAgentBrowserShim(process.env["PATH"])
 		},
+		model,
 		onAbAction: (abAction) => {
 			const action = parseAbAction(abAction);
 			if (action) traceActions.push(action);
@@ -1516,7 +1524,7 @@ ${snapshot}
 }
 //#endregion
 //#region src/diagnose/diagnose.ts
-async function diagnose(input) {
+async function diagnose(input, options = {}) {
 	const { result: raw, isError } = await invokeClaudeStreaming({
 		prompt: buildDiagnosePrompt(input),
 		allowedTools: [
@@ -1524,7 +1532,8 @@ async function diagnose(input) {
 			"Grep",
 			"Glob"
 		],
-		maxTurns: 10
+		maxTurns: 10,
+		model: options.model
 	}, () => {});
 	if (isError) return {
 		result: null,
@@ -1889,7 +1898,7 @@ const DEFAULT_CONFIDENCE_THRESHOLD = .8;
 * or the diagnose loop chose to bail out early.
 */
 async function runAutoFixLoop(input) {
-	const { scriptPath, initialRun, specMarkdown, actions, maxRetries, mode, runVitest, agentBrowserSession, outputLanguage } = input;
+	const { scriptPath, initialRun, specMarkdown, actions, maxRetries, mode, runVitest, agentBrowserSession, outputLanguage, model } = input;
 	let { exitCode, output, currentScript } = initialRun;
 	if (exitCode === 0) return true;
 	for (let attempt = 1; attempt <= maxRetries; attempt++) {
@@ -1905,7 +1914,8 @@ async function runAutoFixLoop(input) {
 			failureLog: output,
 			pageSnapshot: pageSnapshot ?? void 0,
 			mode,
-			outputLanguage
+			outputLanguage,
+			model
 		});
 		if (!fixed) {
 			fix("bailed out; see diagnosis above");
@@ -1920,7 +1930,7 @@ async function runAutoFixLoop(input) {
 	return false;
 }
 async function diagnoseAndFix(input) {
-	const { script, specMarkdown, actions, failureLog, pageSnapshot, mode, outputLanguage } = input;
+	const { script, specMarkdown, actions, failureLog, pageSnapshot, mode, outputLanguage, model } = input;
 	const outcome = await timedPhase("diagnose", () => diagnose({
 		script,
 		specMarkdown,
@@ -1928,11 +1938,11 @@ async function diagnoseAndFix(input) {
 		failureLog,
 		pageSnapshot,
 		outputLanguage
-	}), "fix");
+	}, { model }), "fix");
 	if (outcome.sdkError) {
 		fix("diagnose: SDK error talking to Claude");
 		if (outcome.raw) fix(`diagnose raw: ${truncateForLog(outcome.raw)}`);
-		hint("re-run later, or check ANTHROPIC_API_KEY / network connectivity");
+		hint("re-run later, or check your Claude Code login / network connectivity");
 		return null;
 	}
 	if (!outcome.result) {
@@ -2057,13 +2067,13 @@ function resolveMode(opts) {
 }
 //#endregion
 //#region src/cli/generate.ts
-const generateCommand = new Command("generate").argument("<feature/spec>", "Spec id in '<feature>/<spec>' form (resolves to .ccqa/features/<feature>/test-cases/<spec>/)").description("Generate agent-browser test script from recorded trace actions. test.spec.ts is regenerated from actions.json on every run; pass --force to overwrite manual edits.").option("--max-retries <n>", "Maximum number of auto-fix retries", "3").option("--auto", "Apply auto-fixes without confirmation regardless of confidence (CI use)").option("--no-interactive", "Never prompt; only auto-apply when confidence is high, otherwise give up").option("--force", "Overwrite an existing test.spec.ts without warning").option("--no-snapshot", "Don't pin AGENT_BROWSER_SESSION / capture page snapshots after a failure (debug toggle)").option("--language <bcp47>", "Language for diagnose reasoning / hint text (e.g. 'en', 'ja')", "en").action(async (specPath, opts) => {
+const generateCommand = new Command("generate").argument("<feature/spec>", "Spec id in '<feature>/<spec>' form (resolves to .ccqa/features/<feature>/test-cases/<spec>/)").description("Generate agent-browser test script from recorded trace actions. test.spec.ts is regenerated from actions.json on every run; pass --force to overwrite manual edits.").option("--max-retries <n>", "Maximum number of auto-fix retries", "3").option("--auto", "Apply auto-fixes without confirmation regardless of confidence (CI use)").option("--no-interactive", "Never prompt; only auto-apply when confidence is high, otherwise give up").option("--force", "Overwrite an existing test.spec.ts without warning").option("--no-snapshot", "Don't pin AGENT_BROWSER_SESSION / capture page snapshots after a failure (debug toggle)").option("--language <bcp47>", "Language for diagnose reasoning / hint text (e.g. 'en', 'ja')", "en").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").action(async (specPath, opts) => {
 	const { featureName, specName } = parseSpecPath(specPath);
 	const mode = resolveMode(opts);
 	const useSnapshot = opts.snapshot !== false;
-	await runGenerate(featureName, specName, parseInt(opts.maxRetries, 10), mode, opts.force ?? false, useSnapshot, opts.language ?? "en");
+	await runGenerate(featureName, specName, parseInt(opts.maxRetries, 10), mode, opts.force ?? false, useSnapshot, opts.language ?? "en", opts.model);
 });
-async function runGenerate(featureName, specName, maxRetries, mode, force, useSnapshot, outputLanguage) {
+async function runGenerate(featureName, specName, maxRetries, mode, force, useSnapshot, outputLanguage, model) {
 	header("generate", `${featureName}/${specName}`);
 	await ensureCcqaDir();
 	const existingScriptPath = await getTestScript(featureName, specName);
@@ -2083,7 +2093,7 @@ async function runGenerate(featureName, specName, maxRetries, mode, force, useSn
 	meta("fix-mode", mode);
 	meta("language", outputLanguage);
 	blank();
-	const cleanedActions = await cleanupActions$1(actions);
+	const cleanedActions = await cleanupActions$1(actions, model);
 	if (cleanedActions.length !== actions.length) meta("cleaned", cleanedActions.length);
 	const scriptPath = await saveTestScript(featureName, specName, actionsToScript(cleanedActions, spec.title, setupScripts.length > 0 ? setupScripts : void 0));
 	meta("saved", scriptPath);
@@ -2114,7 +2124,8 @@ async function runGenerate(featureName, specName, maxRetries, mode, force, useSn
 			mode,
 			runVitest: runVitestForSession,
 			agentBrowserSession,
-			outputLanguage
+			outputLanguage,
+			model
 		})) {
 			hint(`run 'ccqa run ${featureName}/${specName}' to execute the test`);
 			return;
@@ -2225,12 +2236,13 @@ async function runVitest$1(scriptPath, agentBrowserSession) {
 		currentScript
 	};
 }
-async function cleanupActions$1(actions) {
+async function cleanupActions$1(actions, model) {
 	try {
 		const { result, isError } = await invokeClaudeStreaming({
 			prompt: buildCleanupPrompt(actions),
 			disableBuiltinTools: true,
-			maxTurns: 1
+			maxTurns: 1,
+			model
 		}, () => {});
 		if (isError || !result) return actions;
 		const json = result.trim().replace(/^```(?:json)?\n?([\s\S]*?)\n?```$/, "$1").trim();
@@ -2409,10 +2421,10 @@ async function resolveSpecs(target) {
 }
 //#endregion
 //#region src/cli/trace-setup.ts
-const traceSetupCommand = new Command("trace-setup").argument("<name>", "Setup name to trace (e.g. login)").description("Trace a setup procedure using dummy placeholder values").action(async (name) => {
-	await runTraceSetup(name);
+const traceSetupCommand = new Command("trace-setup").argument("<name>", "Setup name to trace (e.g. login)").description("Trace a setup procedure using dummy placeholder values").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").action(async (name, opts) => {
+	await runTraceSetup(name, opts.model);
 });
-async function runTraceSetup(name) {
+async function runTraceSetup(name, model) {
 	header("trace-setup", name);
 	await ensureCcqaDir();
 	const spec = parseSetupSpec(await readSetupSpecFile(name));
@@ -2442,6 +2454,7 @@ async function runTraceSetup(name) {
 			PATH: pathWithAgentBrowserShim(process.env["PATH"]),
 			ANTHROPIC_API_KEY: ""
 		},
+		model,
 		onAbAction: (abAction) => {
 			const action = parseAbAction(scrubSecrets(abAction, secretsToScrub));
 			if (action) traceActions.push(action);
@@ -2537,11 +2550,11 @@ function scrubSecrets(line, secrets) {
 }
 //#endregion
 //#region src/cli/generate-setup.ts
-const generateSetupCommand = new Command("generate-setup").argument("<name>", "Setup name to generate (e.g. login)").description("Clean up, validate, and templatize setup actions").option("--max-retries <n>", "Maximum number of auto-fix retries", "3").option("--from-dummy", "Resume from existing test.dummy.spec.ts (after manual fix)").option("--auto", "Apply auto-fixes without confirmation regardless of confidence (CI use)").option("--no-interactive", "Never prompt; only auto-apply when confidence is high, otherwise give up").option("--language <bcp47>", "Language for diagnose reasoning / hint text (e.g. 'en', 'ja')", "en").action(async (name, opts) => {
+const generateSetupCommand = new Command("generate-setup").argument("<name>", "Setup name to generate (e.g. login)").description("Clean up, validate, and templatize setup actions").option("--max-retries <n>", "Maximum number of auto-fix retries", "3").option("--from-dummy", "Resume from existing test.dummy.spec.ts (after manual fix)").option("--auto", "Apply auto-fixes without confirmation regardless of confidence (CI use)").option("--no-interactive", "Never prompt; only auto-apply when confidence is high, otherwise give up").option("--language <bcp47>", "Language for diagnose reasoning / hint text (e.g. 'en', 'ja')", "en").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").action(async (name, opts) => {
 	const mode = resolveMode(opts);
-	await runGenerateSetup(name, parseInt(opts.maxRetries, 10), opts.fromDummy ?? false, mode, opts.language ?? "en");
+	await runGenerateSetup(name, parseInt(opts.maxRetries, 10), opts.fromDummy ?? false, mode, opts.language ?? "en", opts.model);
 });
-async function runGenerateSetup(name, maxRetries, fromDummy, mode, outputLanguage) {
+async function runGenerateSetup(name, maxRetries, fromDummy, mode, outputLanguage, model) {
 	header("generate-setup", name);
 	await ensureCcqaDir();
 	const specContent = await readSetupSpecFile(name);
@@ -2562,7 +2575,7 @@ async function runGenerateSetup(name, maxRetries, fromDummy, mode, outputLanguag
 		meta("fix-mode", mode);
 		meta("language", outputLanguage);
 		blank();
-		cleanedActions = await cleanupActions(actions);
+		cleanedActions = await cleanupActions(actions, model);
 		if (cleanedActions.length !== actions.length) meta("cleaned", cleanedActions.length);
 		await writeFile(dummyPath, actionsToScript(cleanedActions, spec.title), "utf-8");
 		meta("saved", dummyPath);
@@ -2588,7 +2601,8 @@ async function runGenerateSetup(name, maxRetries, fromDummy, mode, outputLanguag
 			mode,
 			runVitest: runVitestForSession,
 			agentBrowserSession,
-			outputLanguage
+			outputLanguage,
+			model
 		});
 		if (!passed) {
 			warn("auto-fix exhausted; setup test still failing");
@@ -2665,12 +2679,13 @@ async function runVitestResolved(scriptPath, agentBrowserSession) {
 		await unlink(tmpPath).catch(() => {});
 	}
 }
-async function cleanupActions(actions) {
+async function cleanupActions(actions, model) {
 	try {
 		const { result, isError } = await invokeClaudeStreaming({
 			prompt: buildCleanupPrompt(actions),
 			disableBuiltinTools: true,
-			maxTurns: 1
+			maxTurns: 1,
+			model
 		}, () => {});
 		if (isError || !result) return actions;
 		const json = result.trim().replace(/^```(?:json)?\n?([\s\S]*?)\n?```$/, "$1").trim();

package/dist/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ccqa",
-  "version": "0.3.7",
+  "version": "0.3.8",
   "type": "module",
   "description": "Browser test recorder powered by Claude Code and agent-browser",
   "repository": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ccqa",
-  "version": "0.3.7",
+  "version": "0.3.8",
   "type": "module",
   "description": "Browser test recorder powered by Claude Code and agent-browser",
   "repository": {