topchester-ai 0.69.0 → 0.71.0

package/dist/bin.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as runTopchesterCli } from "./cli-CUjNKJTW.mjs";
+import { t as runTopchesterCli } from "./cli-BKVU9tvf.mjs";
 //#region src/bin.ts
 await runTopchesterCli();
 //#endregion

package/dist/{cli-CUjNKJTW.mjs → cli-BKVU9tvf.mjs}

@@ -761,7 +761,7 @@ const bashArgsSchema = z.object({
 const bashTool = defineTool({
 	name: "bash",
 	description: "Run an approval-gated shell command inside the workspace.",
-	prompt: "bash: run an approval-gated shell command for terminal work that needs shell syntax, one-off user-requested commands, package manager commands, scripts, pipelines, redirects, or chaining. Prefer run_validator for tests, lint, typecheck, build, check, format-check, and smoke. To use it, reply with only JSON: {\"tool\":\"bash\",\"args\":{\"command\":\"printf hi | wc -c\",\"workdir\":\".\",\"timeout_ms\":120000,\"description\":\"count bytes\"}}",
+	prompt: "bash: run an approval-gated shell command for terminal work that needs shell syntax, one-off user-requested commands, package manager commands, scripts, pipelines, redirects, or chaining. Use run_validator, not bash, for tests and checks that fit strict validator shapes such as pnpm test, go test, cargo test, node --test, local npx tsx --test, lint, typecheck, build, check, format-check, and smoke. To use it, reply with only JSON: {\"tool\":\"bash\",\"args\":{\"command\":\"printf hi | wc -c\",\"workdir\":\".\",\"timeout_ms\":120000,\"description\":\"count bytes\"}}",
 	argsSchema: bashArgsSchema,
 	requiresExclusiveWorkspace: true,
 	execute: async (context, args) => runBashCommand(context.workspaceRoot, args, {
@@ -4355,8 +4355,6 @@ const DANGEROUS_EXECUTABLES = new Set([
 	"git",
 	"kubectl",
 	"mv",
-	"npx",
-	"pnpx",
 	"rm",
 	"rmdir",
 	"scp",
@@ -4585,13 +4583,24 @@ async function findNearestPackageMetadata(workspaceRoot, cwd) {
 async function readPackageMetadata(path) {
 	const raw = JSON.parse(await readFile(path, "utf8"));
 	const scripts = Object.fromEntries(Object.entries(raw.scripts ?? {}).filter((entry) => typeof entry[1] === "string"));
+	const dependencies = new Set([
+		...getPackageDependencyNames(raw.dependencies),
+		...getPackageDependencyNames(raw.devDependencies),
+		...getPackageDependencyNames(raw.optionalDependencies),
+		...getPackageDependencyNames(raw.peerDependencies)
+	]);
 	return {
 		path,
 		dir: dirname(path),
 		scripts,
+		dependencies,
 		packageManager: parsePackageManager(raw.packageManager) ?? await detectPackageManagerFromLockfiles(dirname(path))
 	};
 }
+function getPackageDependencyNames(value) {
+	if (!value || typeof value !== "object" || Array.isArray(value)) return [];
+	return Object.keys(value);
+}
 function parsePackageManager(value) {
 	if (typeof value !== "string") return;
 	const name = value.split("@")[0];
@@ -4614,8 +4623,25 @@ async function detectPackageManagerFromLockfiles(dir) {
 }
 function classifyValidator(command, metadata) {
 	if (isPackageManager(command.executable)) return classifyPackageManagerCommand(command, metadata);
+	if (command.executable === "npx" || command.executable === "pnpx") return classifyPackageExecutorCommand(command, metadata);
 	return classifyDirectValidator(command);
 }
+function classifyPackageExecutorCommand(command, metadata) {
+	const args = command.args[0] === "--no-install" ? command.args.slice(1) : command.args;
+	const executable = args[0];
+	if (executable !== "tsx") return {
+		allowed: false,
+		reason: `command policy rejected '${command.executable}' because only local tsx validators are allowed.`
+	};
+	if (!metadata?.dependencies.has("tsx")) return {
+		allowed: false,
+		reason: `command policy rejected '${command.executable} tsx' because package.json does not declare tsx.`
+	};
+	return classifyDirectValidator({
+		executable,
+		args: args.slice(1)
+	});
+}
 function classifyPackageManagerCommand(command, metadata) {
 	const manager = command.executable;
 	const firstArg = command.args[0];
@@ -4689,6 +4715,15 @@ function getPackageScriptName(manager, args) {
 }
 function classifyDirectValidator(command) {
 	switch (command.executable) {
+		case "go": return classifyGoValidator(command);
+		case "cargo": return classifyCargoValidator(command);
+		case "tsx": return command.args[0] === "--test" || command.args.includes("--test") ? {
+			allowed: true,
+			validator: "test"
+		} : {
+			allowed: false,
+			reason: "command policy rejected 'tsx' because only 'tsx --test' is a validator."
+		};
 		case "vitest":
 		case "jest":
 		case "mocha": return {
@@ -4739,6 +4774,30 @@ function classifyDirectValidator(command) {
 		reason: `command policy rejected '${command.executable}' because it is not a known validator.`
 	};
 }
+function classifyGoValidator(command) {
+	if (command.args[0] !== "test") return {
+		allowed: false,
+		reason: "command policy rejected 'go' because only 'go test' is a validator."
+	};
+	if (command.args.includes("-c") || command.args.some((arg) => arg.startsWith("-exec") || arg.startsWith("-toolexec"))) return {
+		allowed: false,
+		reason: "command policy rejected 'go test' because compile-only or custom tool execution flags are not allowed."
+	};
+	return {
+		allowed: true,
+		validator: "test"
+	};
+}
+function classifyCargoValidator(command) {
+	if (command.args[0] !== "test") return {
+		allowed: false,
+		reason: "command policy rejected 'cargo' because only 'cargo test' is a validator."
+	};
+	return {
+		allowed: true,
+		validator: "test"
+	};
+}
 function classifyScriptName(scriptName) {
 	if (scriptName === "test" || scriptName.startsWith("test:")) return "test";
 	if (scriptName === "lint" || scriptName.startsWith("lint:")) return "lint";
@@ -4771,7 +4830,7 @@ const validatorKindSchema = z.preprocess((value) => value === "format" || value
 const runValidatorTool = defineTool({
 	name: "run_validator",
 	description: "Run a strictly validated test, lint, typecheck, build, check, or smoke command inside the workspace.",
-	prompt: "run_validator: run a strict verification command after edits, such as tests, lint, typecheck, build, check, format-check, or smoke; format means check-only commands such as pnpm format-check, not mutating formatter commands such as pnpm format; failed exits are useful evidence and should be inspected before retrying. To use it, reply with only JSON: {\"tool\":\"run_validator\",\"args\":{\"command\":\"pnpm test test/tools.test.ts\",\"validator\":\"test\",\"workdir\":\".\",\"timeout_ms\":120000}}",
+	prompt: "run_validator: run a strict verification command after edits, such as pnpm test, go test ./..., cargo test, node --test, local npx tsx --test, lint, typecheck, build, check, format-check, or smoke; format means check-only commands such as pnpm format-check, not mutating formatter commands such as pnpm format; failed exits are useful evidence and should be inspected before retrying. To use it, reply with only JSON: {\"tool\":\"run_validator\",\"args\":{\"command\":\"pnpm test test/tools.test.ts\",\"validator\":\"test\",\"workdir\":\".\",\"timeout_ms\":120000}}",
 	argsSchema: z.object({
 		command: z.string().min(1).max(2e3),
 		validator: validatorKindSchema.optional(),
@@ -5287,8 +5346,8 @@ function formatSkillSource$2(skill) {
 }
 const taskTool = defineTool({
 	name: "task",
-	description: "Delegate read-only file/search/git research to a child agent. Do not use for shell commands, bash, Python/Node scripts, validators, edits, writes, or other execution work.",
-	prompt: "task: delegate read-only file/search/git research to a child agent. Do not use task for shell commands, bash, Python/Node scripts, validators, edits, writes, finish_task, or other execution work; use parent tools directly. To use it, reply with only JSON: {\"tool\":\"task\",\"args\":{\"description\":\"Inspect runtime event flow\",\"prompt\":\"Read the runtime and summarize how events are emitted.\",\"subagent_type\":\"explore\"}}",
+	description: "Delegate read-only file/search/git research to a child agent. Do not use for shell commands, bash, Python/Node scripts, validators, edits, writes, tiny local inspections, or other execution work.",
+	prompt: "task: delegate read-only file/search/git research to a child agent. Do not use task for shell commands, bash, Python/Node scripts, validators, edits, writes, finish_task, tiny local inspections, or other execution work; use parent tools directly. If the relevant workspace context is just a README plus a few obvious source files, inspect them directly with list_files/read_file instead of spawning a subagent. To use it, reply with only JSON: {\"tool\":\"task\",\"args\":{\"description\":\"Inspect runtime event flow\",\"prompt\":\"Read the runtime and summarize how events are emitted.\",\"subagent_type\":\"explore\"}}",
 	argsSchema: z.object({
 		description: z.string().min(1),
 		prompt: z.string().min(1),
@@ -12666,6 +12725,7 @@ function getChatSystemPrompt(options = {}) {
 		] : [],
 		...canUseTool("inspect_command") ? ["- Do not use inspect_command for file creation or file mutation."] : [],
 		...canUseTool("edit_file") ? ["- Keep edit_file old_text small but unique. Do not include line labels or grep prefixes in old_text; use exact file text only."] : [],
+		...canUseTool("apply_patch") && canUseTool("write_file") && canUseTool("read_file") ? ["- If apply_patch/edit_file fails twice on the same file, stop retrying equivalent patches. Read the current file and, when the intended change is broad, use write_file overwrite:true with the current read_file hash to replace the whole file."] : [],
 		...canUseTool("edit_file") || canUseTool("write_file") ? ["- Use edit/write/patch tools when they are available and the user asks you to implement, fix, add, update, or refactor code."] : [],
 		...canUseTool("finish_task") ? ["- Use finish_task to complete implementation tasks after the requested work is actually done. Do not call finish_task to claim edits or validation that tool results did not confirm."] : [],
 		...canUseTool("inspect_command") || canUseTool("run_validator") ? ["- Use command/test tools when they are available and you need to inspect the environment or verify behavior."] : [],
@@ -16787,4 +16847,4 @@ function formatDryRunSyncStatus(status) {
 //#endregion
 export { runTopchesterCli as t };
 
-//# sourceMappingURL=cli-CUjNKJTW.mjs.map
+//# sourceMappingURL=cli-BKVU9tvf.mjs.map