@bjoernrochel/sevro 0.1.0-rc.7 → 1.0.0

package/src/hosts/codex-exec-controls.ts

@@ -0,0 +1,516 @@
+import { createRequire } from "node:module";
+import type ts from "typescript";
+import { isRecord, isUnknownArray } from "../value-guards";
+
+const require = createRequire(import.meta.url);
+let parser: typeof ts | undefined;
+function typescript(): typeof ts {
+  return (parser ??= require("typescript") as typeof ts);
+}
+
+interface Entry {
+  ordinal: number;
+  payload: Record<string, unknown>;
+}
+interface SubmittedCall {
+  index: number;
+  namespace: "functions" | "collaboration" | "other";
+  name: string;
+}
+export interface ExecCallReceipt {
+  ordinal: number;
+  completeness: "complete" | "partial";
+  calls: SubmittedCall[];
+  inspectionReason?:
+    | "unsupported_source"
+    | "malformed_source"
+    | "source_limit"
+    | "foreign_namespace"
+    | null;
+  execution?: "completed" | "failed" | "incomplete";
+  output?: "complete" | "truncated" | "unavailable";
+}
+
+function toolNamespace(name: string): SubmittedCall["namespace"] {
+  if (
+    [
+      "spawn_agent",
+      "list_agents",
+      "wait_agent",
+      "followup_task",
+      "send_message",
+      "interrupt_agent",
+    ].includes(name)
+  )
+    return "collaboration";
+  if (
+    [
+      "create_goal",
+      "get_goal",
+      "update_goal",
+      "exec_command",
+      "write_stdin",
+    ].includes(name)
+  )
+    return "functions";
+  return "other";
+}
+function toolName(node: ts.Expression): string | undefined {
+  const ts = typescript();
+  if (!ts.isPropertyAccessExpression(node)) return undefined;
+  if (!ts.isIdentifier(node.expression) || node.expression.text !== "tools")
+    return undefined;
+  return /^[A-Za-z_][A-Za-z0-9_]{0,63}$/.test(node.name.text)
+    ? node.name.text
+    : undefined;
+}
+function literal(node: ts.Expression): boolean {
+  const ts = typescript();
+  if (ts.isStringLiteralLike(node) || ts.isNumericLiteral(node)) return true;
+  return [
+    ts.SyntaxKind.TrueKeyword,
+    ts.SyntaxKind.FalseKeyword,
+    ts.SyntaxKind.NullKeyword,
+  ].includes(node.kind);
+}
+function member(node: ts.Expression, owner: string, names: string[]): boolean {
+  const ts = typescript();
+  return (
+    ts.isPropertyAccessExpression(node) &&
+    ts.isIdentifier(node.expression) &&
+    node.expression.text === owner &&
+    names.includes(node.name.text)
+  );
+}
+function localName(node: ts.BindingName): string | undefined {
+  const ts = typescript();
+  if (!ts.isIdentifier(node)) return undefined;
+  return [
+    "tools",
+    "text",
+    "store",
+    "Promise",
+    "load",
+    "JSON",
+    "String",
+  ].includes(node.text)
+    ? undefined
+    : node.text;
+}
+class Submission {
+  readonly calls: SubmittedCall[] = [];
+  private readonly locals = new Set<string>();
+  private readonly collections = new Set<string>();
+  private readonly indices = new Set<string>();
+  private nodes = 0;
+  expression(node: ts.Expression): boolean {
+    const ts = typescript();
+    if (++this.nodes > 2048) return false;
+    switch (true) {
+      case literal(node):
+        return true;
+      case ts.isTemplateExpression(node):
+        return node.templateSpans.every((span) =>
+          this.expression(span.expression),
+        );
+      case ts.isBinaryExpression(node):
+        return this.composition(node);
+      case ts.isAwaitExpression(node):
+        return this.expression(node.expression);
+      case ts.isObjectLiteralExpression(node):
+        return node.properties.every((property) => this.property(property));
+      case ts.isArrayLiteralExpression(node):
+        return node.elements.every((element) => this.expression(element));
+      case ts.isIdentifier(node):
+        return this.locals.has(node.text);
+      case ts.isPropertyAccessExpression(node):
+        return this.localProperty(node);
+      case ts.isCallExpression(node):
+        return this.call(node);
+      case ts.isConditionalExpression(node):
+        return [node.condition, node.whenTrue, node.whenFalse].every((part) =>
+          this.outputExpression(part),
+        );
+      case ts.isElementAccessExpression(node):
+        return (
+          this.collection(node.expression) &&
+          typescript().isIdentifier(node.argumentExpression) &&
+          this.indices.has(node.argumentExpression.text)
+        );
+      default:
+        return false;
+    }
+  }
+  private property(node: ts.ObjectLiteralElementLike): boolean {
+    const ts = typescript();
+    if (ts.isSpreadAssignment(node))
+      return this.outputExpression(node.expression);
+    if (ts.isShorthandPropertyAssignment(node)) return this.shorthand(node);
+    if (!ts.isPropertyAssignment(node) || ts.isComputedPropertyName(node.name))
+      return false;
+    return this.expression(node.initializer);
+  }
+  private shorthand(node: ts.ShorthandPropertyAssignment): boolean {
+    return !node.objectAssignmentInitializer && this.expression(node.name);
+  }
+  private composition(node: ts.BinaryExpression): boolean {
+    if (
+      [
+        typescript().SyntaxKind.EqualsEqualsEqualsToken,
+        typescript().SyntaxKind.ExclamationEqualsEqualsToken,
+      ].includes(node.operatorToken.kind)
+    )
+      return (
+        this.outputExpression(node.left) && this.outputExpression(node.right)
+      );
+    return (
+      node.operatorToken.kind === typescript().SyntaxKind.PlusToken &&
+      this.expression(node.left) &&
+      this.expression(node.right)
+    );
+  }
+  private call(node: ts.CallExpression): boolean {
+    if (this.outputEach(node)) return true;
+    if (!node.arguments.every((argument) => this.expression(argument)))
+      return false;
+    return helperCall(node.expression) || this.retainToolCall(node.expression);
+  }
+  private retainToolCall(expression: ts.Expression): boolean {
+    const name = toolName(expression);
+    if (!name || this.calls.length >= 128) return false;
+    this.calls.push({
+      index: this.calls.length,
+      namespace: toolNamespace(name),
+      name,
+    });
+    return name !== "exec";
+  }
+  statement(node: ts.Statement): boolean {
+    const ts = typescript();
+    if (ts.isVariableStatement(node))
+      return this.declarations(node.declarationList);
+    if (ts.isForOfStatement(node)) return this.outputLoop(node);
+    if (ts.isForStatement(node)) return this.numericOutputLoop(node);
+    return ts.isExpressionStatement(node) && this.expression(node.expression);
+  }
+  private localProperty(node: ts.PropertyAccessExpression): boolean {
+    return (
+      this.expression(node.expression) &&
+      !["constructor", "__proto__", "prototype"].includes(node.name.text)
+    );
+  }
+  private declarations(list: ts.VariableDeclarationList): boolean {
+    const ts = typescript();
+    if (!(list.flags & ts.NodeFlags.Const)) return false;
+    return list.declarations.every((declaration) =>
+      this.declaration(declaration),
+    );
+  }
+  private declaration(node: ts.VariableDeclaration): boolean {
+    if (!node.initializer) return false;
+    const names = this.bindingNames(node.name, node.initializer);
+    if (!names) return false;
+    if (!this.expression(node.initializer)) return false;
+    names.forEach((name) => this.locals.add(name));
+    this.retainCollection(node.name, node.initializer, names);
+    return true;
+  }
+  private retainCollection(
+    name: ts.BindingName,
+    initializer: ts.Expression,
+    names: string[],
+  ): void {
+    if (localName(name) && collectionExpression(initializer))
+      names.forEach((name) => this.collections.add(name));
+  }
+  private bindingNames(
+    node: ts.BindingName,
+    initializer: ts.Expression,
+  ): string[] | undefined {
+    const ts = typescript();
+    if (ts.isIdentifier(node)) {
+      const name = this.declaredName(node);
+      return name ? [name] : undefined;
+    }
+    if (!ts.isArrayBindingPattern(node) || !collectionExpression(initializer))
+      return undefined;
+    return this.batchNames(node);
+  }
+  private batchNames(node: ts.ArrayBindingPattern): string[] | undefined {
+    const names = node.elements.map((element) => this.batchBinding(element));
+    if (!names.every((name): name is string => name !== undefined))
+      return undefined;
+    return new Set(names).size === names.length ? names : undefined;
+  }
+  private batchBinding(node: ts.ArrayBindingElement): string | undefined {
+    const ts = typescript();
+    if (!ts.isBindingElement(node)) return undefined;
+    if (node.dotDotDotToken || node.initializer || node.propertyName)
+      return undefined;
+    return this.declaredName(node.name);
+  }
+  private declaredName(node: ts.BindingName): string | undefined {
+    const name = localName(node);
+    return name && !this.locals.has(name) ? name : undefined;
+  }
+  private outputEach(node: ts.CallExpression): boolean {
+    const ts = typescript();
+    const [callback] = node.arguments;
+    if (
+      node.arguments.length !== 1 ||
+      !callback ||
+      !ts.isIdentifier(callback) ||
+      callback.text !== "text"
+    )
+      return false;
+    return [...this.collections].some((name) =>
+      member(node.expression, name, ["forEach"]),
+    );
+  }
+  private outputLoop(node: ts.ForOfStatement): boolean {
+    const ts = typescript();
+    if (!this.collection(node.expression)) return false;
+    if (!ts.isVariableDeclarationList(node.initializer)) return false;
+    const name = loopVariable(node.initializer);
+    if (!name || this.locals.has(name)) return false;
+    this.locals.add(name);
+    const valid = this.outputBody(node.statement);
+    this.locals.delete(name);
+    return valid;
+  }
+  private outputExpression(node: ts.Expression): boolean {
+    const count = this.calls.length;
+    return this.expression(node) && this.calls.length === count;
+  }
+  private numericOutputLoop(node: ts.ForStatement): boolean {
+    const name = numericLoopVariable(node);
+    if (!name || this.locals.has(name)) return false;
+    const condition = node.condition as ts.BinaryExpression;
+    const bound = condition.right as ts.PropertyAccessExpression;
+    if (!this.collection(bound.expression)) return false;
+    this.locals.add(name);
+    this.indices.add(name);
+    const valid = this.outputBody(node.statement);
+    this.locals.delete(name);
+    this.indices.delete(name);
+    return valid;
+  }
+  private outputBody(node: ts.Statement): boolean {
+    const ts = typescript();
+    if (ts.isBlock(node))
+      return node.statements.every((statement) => this.outputBody(statement));
+    if (ts.isIfStatement(node)) return this.outputConditional(node);
+    return (
+      ts.isExpressionStatement(node) && this.outputExpression(node.expression)
+    );
+  }
+  private outputConditional(node: ts.IfStatement): boolean {
+    return (
+      this.outputExpression(node.expression) &&
+      this.outputBody(node.thenStatement) &&
+      (!node.elseStatement || this.outputBody(node.elseStatement))
+    );
+  }
+  private collection(node: ts.Expression): boolean {
+    const ts = typescript();
+    return ts.isIdentifier(node) && this.collections.has(node.text);
+  }
+}
+function helperCall(node: ts.Expression): boolean {
+  const ts = typescript();
+  if (ts.isIdentifier(node) && ["text", "store", "String"].includes(node.text))
+    return true;
+  return (
+    member(node, "Promise", ["all", "allSettled"]) ||
+    member(node, "JSON", ["stringify"])
+  );
+}
+function namedIdentifier(
+  node: ts.Expression | undefined,
+  name: string,
+): boolean {
+  return (
+    node !== undefined && typescript().isIdentifier(node) && node.text === name
+  );
+}
+function collectionExpression(node: ts.Expression): boolean {
+  const ts = typescript();
+  if (!ts.isAwaitExpression(node) || !ts.isCallExpression(node.expression))
+    return false;
+  return member(node.expression.expression, "Promise", ["all", "allSettled"]);
+}
+function loopVariable(list: ts.VariableDeclarationList): string | undefined {
+  const ts = typescript();
+  if (!(list.flags & ts.NodeFlags.Const) || list.declarations.length !== 1)
+    return undefined;
+  const declaration = list.declarations[0];
+  return declaration && !declaration.initializer
+    ? localName(declaration.name)
+    : undefined;
+}
+function numericLoopVariable(node: ts.ForStatement): string | undefined {
+  const ts = typescript();
+  if (!node.initializer || !ts.isVariableDeclarationList(node.initializer))
+    return undefined;
+  const name = numericBinding(node.initializer);
+  return name && numericLoopBounds(node, name) ? name : undefined;
+}
+function numericBinding(list: ts.VariableDeclarationList): string | undefined {
+  if (
+    !(list.flags & typescript().NodeFlags.Let) ||
+    list.declarations.length !== 1
+  )
+    return undefined;
+  const declaration = list.declarations[0];
+  return declaration && zeroInitializer(declaration.initializer)
+    ? localName(declaration.name)
+    : undefined;
+}
+function zeroInitializer(node: ts.Expression | undefined): boolean {
+  return (
+    node !== undefined &&
+    typescript().isNumericLiteral(node) &&
+    node.text === "0"
+  );
+}
+function numericLoopBounds(node: ts.ForStatement, name: string): boolean {
+  const ts = typescript();
+  const condition = node.condition;
+  if (!condition || !ts.isBinaryExpression(condition)) return false;
+  return (
+    numericCondition(condition, name) &&
+    numericIncrement(node.incrementor, name)
+  );
+}
+function numericCondition(node: ts.BinaryExpression, name: string): boolean {
+  return (
+    node.operatorToken.kind === typescript().SyntaxKind.LessThanToken &&
+    namedIdentifier(node.left, name) &&
+    lengthProperty(node.right)
+  );
+}
+function lengthProperty(node: ts.Expression): boolean {
+  return (
+    typescript().isPropertyAccessExpression(node) && node.name.text === "length"
+  );
+}
+function numericIncrement(
+  node: ts.Expression | undefined,
+  name: string,
+): boolean {
+  return (
+    node !== undefined &&
+    typescript().isPostfixUnaryExpression(node) &&
+    node.operator === typescript().SyntaxKind.PlusPlusToken &&
+    namedIdentifier(node.operand, name)
+  );
+}
+function inspectedSource(code: unknown): {
+  submission?: Submission;
+  reason: ExecCallReceipt["inspectionReason"];
+} {
+  if (typeof code !== "string") return { reason: "unsupported_source" };
+  if (code.length > 65536) return { reason: "source_limit" };
+  const ts = typescript();
+  const source = ts.createSourceFile(
+    "submitted.js",
+    code,
+    ts.ScriptTarget.Latest,
+    false,
+    ts.ScriptKind.JS,
+  );
+  const diagnostics = (
+    source as ts.SourceFile & { parseDiagnostics: readonly ts.Diagnostic[] }
+  ).parseDiagnostics;
+  if (diagnostics.length !== 0) return { reason: "malformed_source" };
+  const submission = new Submission();
+  return source.statements.every((statement) => submission.statement(statement))
+    ? { submission, reason: null }
+    : { reason: "unsupported_source" };
+}
+function outputBlocks(output: unknown): unknown[] {
+  if (isUnknownArray(output)) return output;
+  if (typeof output !== "string") return [];
+  try {
+    const blocks: unknown = JSON.parse(output);
+    return isUnknownArray(blocks) ? blocks : [];
+  } catch {
+    return [];
+  }
+}
+function completeTextBlock(block: unknown): boolean {
+  if (!isRecord(block) || !["text", "input_text"].includes(String(block.type)))
+    return false;
+  return typeof block.text === "string";
+}
+function completedHeader(block: unknown): boolean {
+  return (
+    isRecord(block) &&
+    typeof block.text === "string" &&
+    /^Script completed\nWall time [0-9]+(?:\.[0-9]+)? seconds\nOutput:\n$/.test(
+      block.text,
+    )
+  );
+}
+function submissionOutput(entry: Entry, entries: Entry[]): unknown[] {
+  const id = entry.payload.call_id;
+  if (typeof id !== "string") return [];
+  const matches = entries.filter((other) => other.payload.call_id === id);
+  if (matches.length !== 2) return [];
+  const result = matches.find(
+    (other) => other.payload.type === "custom_tool_call_output",
+  );
+  return result && result.ordinal > entry.ordinal
+    ? outputBlocks(result.payload.output)
+    : [];
+}
+function outputEvidence(
+  blocks: unknown[],
+): Pick<ExecCallReceipt, "execution" | "output"> {
+  if (!blocks.length || !blocks.every(completeTextBlock))
+    return { execution: "incomplete", output: "unavailable" };
+  const header = blocks[0] as { text: string };
+  const execution = executionStatus(header);
+  const truncated = blocks.some((block) =>
+    (block as { text: string }).text.includes("Warning: truncated output"),
+  );
+  return { execution, output: truncated ? "truncated" : "complete" };
+}
+function executionStatus(header: {
+  text: string;
+}): ExecCallReceipt["execution"] {
+  if (completedHeader(header)) return "completed";
+  return /^Script failed\b/.test(header.text) ? "failed" : "incomplete";
+}
+function nativeExecNamespace(namespace: unknown): boolean {
+  return namespace === undefined || namespace === "functions";
+}
+/** Static inspection only: submitted code is never executed by the observer. */
+export function codexExecCallReceipts(entries: Entry[]): ExecCallReceipt[] {
+  return entries
+    .filter(
+      ({ payload }) =>
+        payload.type === "custom_tool_call" && payload.name === "exec",
+    )
+    .map((entry) => ({
+      ordinal: entry.ordinal,
+      ...inspectionEvidence(entry),
+      ...outputEvidence(submissionOutput(entry, entries)),
+    }));
+}
+function inspectionEvidence(
+  entry: Entry,
+): Pick<ExecCallReceipt, "completeness" | "calls" | "inspectionReason"> {
+  if (!nativeExecNamespace(entry.payload.namespace))
+    return {
+      completeness: "partial",
+      calls: [],
+      inspectionReason: "foreign_namespace",
+    };
+  const inspected = inspectedSource(entry.payload.input);
+  return {
+    completeness: inspected.submission ? "complete" : "partial",
+    calls: inspected.submission?.calls ?? [],
+    inspectionReason: inspected.reason,
+  };
+}

package/src/hosts/codex-marketplace.ts

@@ -1,10 +1,12 @@
 import type { Dirent } from "node:fs";
 import { readFile, readdir, realpath, stat } from "node:fs/promises";
-import { isAbsolute, join, relative, sep } from "node:path";
-import { fixtureParts } from "../preparation";
+import { join, relative, sep } from "node:path";
+import { insideRoot as inside } from "../paths";
+import { fixtureParts } from "../fixtures/preparation";
 import type { HostAdapter } from "../engine";
 import { runCodexProcess } from "./codex-process";
 import { isRecord, isUnknownArray } from "../value-guards";
+import type { HostDiagnostics } from "../host-diagnostics";
 type Request = Parameters<HostAdapter["run"]>[0];
 type Declaration = NonNullable<Request["codexMarketplace"]>;
 interface InstallContext {
@@ -12,15 +14,9 @@ interface InstallContext {
   env: Record<string, string>;
   pluginCacheRoot: string;
   request: Request;
+  diagnostics: HostDiagnostics;
 }
 
-function inside(root: string, path: string): boolean {
-  const child = relative(root, path);
-  return (
-    child === "" ||
-    (child !== ".." && !child.startsWith(`..${sep}`) && !isAbsolute(child))
-  );
-}
 function validPluginNames(declaration: Declaration): boolean {
   const names = declaration.pluginNames;
   return (
@@ -174,20 +170,24 @@ function verifyPackageEntry(
 }
 
 async function pluginCommand(context: InstallContext, args: string[]) {
-  return runCodexProcess({
+  const result = await runCodexProcess({
+    diagnostics: context.diagnostics,
     argv: [context.binary, "plugin", ...args, "--json"],
     cwd: context.request.workspace,
     env: context.env,
     timeoutMs: 30_000,
     signal: context.request.signal,
   });
+  return result;
 }
 export async function installCodexPlugins(
   context: InstallContext,
 ): Promise<string[]> {
   const declaration = context.request.codexMarketplace;
   if (!declaration) return [];
+  context.diagnostics.stage = "marketplace-validation";
   const root = await marketplaceRoot(context.request.workspace, declaration);
+  context.diagnostics.stage = "marketplace-install";
   const added = await pluginCommand(context, ["marketplace", "add", root]);
   if (added.code !== 0)
     throw new Error("Codex local marketplace installation failed");
@@ -201,12 +201,14 @@ async function installPlugin(
   declaration: Declaration,
   pluginName: string,
 ): Promise<string> {
+  context.diagnostics.stage = "plugin-install";
   const installed = await pluginCommand(context, [
     "add",
     `${pluginName}@${declaration.marketplaceName}`,
   ]);
   if (installed.code !== 0)
     throw new Error("Codex local plugin installation failed");
+  context.diagnostics.stage = "plugin-receipt";
   const path = await installedPluginPath(
     installed.out,
     declaration.marketplaceName,