support merging from codex, fix docker (#234)

agents/codex.ts

@@ -43,17 +43,23 @@ function writeCodexConfig(ctx: AgentRunContext): string {
   }
   const featuresSection = features.length > 0 ? `[features]\n${features.join("\n")}` : "";
 
+  // trust the project so codex loads repo-level .codex/config.toml
+  const cwd = process.cwd();
+  const projectTrustSection = `[projects."${cwd}"]\ntrust_level = "trusted"`;
+
   writeFileSync(
     configPath,
     `# written by pullfrog
 ${featuresSection}
 
+${projectTrustSection}
+
 ${mcpServerSections.join("\n\n")}
 `.trim() + "\n"
   );
 
   log.info(
-    `» Codex config written to ${configPath} (shell: ${bash === "enabled" ? "enabled" : "disabled"})`
+    `» Codex config written to ${configPath} (shell: ${bash === "enabled" ? "enabled" : "disabled"}, project trusted: ${cwd})`
   );
 
   return codexDir;
@@ -133,7 +139,6 @@ export const codex = agent({
 
     const env: NodeJS.ProcessEnv = {
       ...process.env,
-      HOME: ctx.tmpdir,
       CODEX_HOME: codexDir,
       CODEX_API_KEY: apiKey,
     };

entry

@@ -144457,16 +144457,21 @@ url = "${ctx.mcpServerUrl}"`];
   }
   const featuresSection = features.length > 0 ? `[features]
 ${features.join("\n")}` : "";
+  const cwd = process.cwd();
+  const projectTrustSection = `[projects."${cwd}"]
+trust_level = "trusted"`;
   writeFileSync7(
     configPath,
     `# written by pullfrog
 ${featuresSection}
 
+${projectTrustSection}
+
 ${mcpServerSections.join("\n\n")}
 `.trim() + "\n"
   );
   log.info(
-    `\xBB Codex config written to ${configPath} (shell: ${bash === "enabled" ? "enabled" : "disabled"})`
+    `\xBB Codex config written to ${configPath} (shell: ${bash === "enabled" ? "enabled" : "disabled"}, project trusted: ${cwd})`
   );
   return codexDir;
 }
@@ -144522,7 +144527,6 @@ var codex = agent({
     const commandExecutionIds = /* @__PURE__ */ new Set();
     const env3 = {
       ...process.env,
-      HOME: ctx.tmpdir,
       CODEX_HOME: codexDir,
       CODEX_API_KEY: apiKey
     };

test/crossagent/mcpmerge.ts

@@ -2,27 +2,15 @@ import type { AgentResult, TestRunnerOptions, ValidationCheck } from "../utils.t
 import { defineFixture, generateAgentUuids, getStructuredOutput } from "../utils.ts";
 
 /**
- * MCP merge tests - validate MCP server configuration with repo-level MCP configs present.
+ * MCP merge test - validates repo-level MCP servers merge correctly with gh_pullfrog.
  *
  * uses GITHUB_REPOSITORY=pullfrog/test-repo-mcp which has robinMCP server.
- *
+ * all agents should auto-discover repo-level MCP configs and merge them with gh_pullfrog.
- * two variants:
- * - mcpmerge-full: validates repo-level MCP servers merge correctly with gh_pullfrog
- *   (claude, cursor, gemini, opencode - agents that auto-discover repo MCPs)
- * - mcpmerge-pullfrog-only: validates gh_pullfrog remains available when repo has MCP config
- *   (codex - doesn't auto-discover repo MCPs)
  */
 
-// shared env for both tests
+const testUuids = generateAgentUuids(["PULLFROG_MCP_TEST"]);
-const sharedEnv = {
-  GITHUB_REPOSITORY: "pullfrog/test-repo-mcp",
-};
 
-// --- mcpmerge-full: test repo-level MCP discovery ---
+const fixture = defineFixture(
-
-const fullTestUuids = generateAgentUuids(["PULLFROG_MCP_TEST"]);
-
-const fullFixture = defineFixture(
   {
     prompt: `Call the get_test_value tool from the robinMCP server, then call set_output with the exact value returned.`,
     effort: "mini",
@@ -30,10 +18,10 @@ const fullFixture = defineFixture(
   { localOnly: true }
 );
 
-function fullValidator(result: AgentResult): ValidationCheck[] {
+function validator(result: AgentResult): ValidationCheck[] {
   const output = getStructuredOutput(result);
   const setOutputCalled = output !== null;
-  const expectedUuid = fullTestUuids.getUuid(result.agent, "PULLFROG_MCP_TEST");
+  const expectedUuid = testUuids.getUuid(result.agent, "PULLFROG_MCP_TEST");
   const correctValue = setOutputCalled && output === expectedUuid;
 
   return [
@@ -42,46 +30,12 @@ function fullValidator(result: AgentResult): ValidationCheck[] {
   ];
 }
 
-// --- mcpmerge-pullfrog-only: test gh_pullfrog availability only ---
+export const test: TestRunnerOptions = {
-
+  name: "mcpmerge",
-const baseFixture = defineFixture(
+  fixture,
-  {
+  validator,
-    prompt: `Call set_output with "PULLFROG_MCP_WORKS".`,
+  tags: ["mcpmerge"],
-    effort: "mini",
+  env: { GITHUB_REPOSITORY: "pullfrog/test-repo-mcp" },
-  },
+  agentEnv: testUuids.agentEnv,
-  { localOnly: true }
+  fileAgentEnv: testUuids.agentEnv,
-);
-
-function baseValidator(result: AgentResult): ValidationCheck[] {
-  const output = getStructuredOutput(result);
-  const setOutputCalled = output !== null;
-  const correctValue = setOutputCalled && output === "PULLFROG_MCP_WORKS";
-
-  return [
-    { name: "set_output", passed: setOutputCalled },
-    { name: "correct_value", passed: correctValue },
-  ];
-}
-
-// --- exports ---
-
-export const tests: Record<string, TestRunnerOptions> = {
-  "mcpmerge-full": {
-    name: "mcpmerge-full",
-    fixture: fullFixture,
-    validator: fullValidator,
-    tags: ["mcpmerge"],
-    agents: ["claude", "cursor", "gemini", "opencode"],
-    env: sharedEnv,
-    agentEnv: fullTestUuids.agentEnv,
-    fileAgentEnv: fullTestUuids.agentEnv,
-  },
-  "mcpmerge-pullfrog-only": {
-    name: "mcpmerge-pullfrog-only",
-    fixture: baseFixture,
-    validator: baseValidator,
-    tags: ["mcpmerge"],
-    agents: ["codex"],
-    env: sharedEnv,
-  },
 };

utils/docker.ts

@@ -202,6 +202,9 @@ export function buildDockerRunArgs(config: DockerRunArgsContext): string[] {
   const setupCmd = [
     // install sudo (node:24 is Debian-based) - check if already installed first
     `which sudo > /dev/null 2>&1 || (apt-get update -qq && apt-get install -qq -y sudo > /dev/null 2>&1)`,
+    // remove any existing user/group with the same uid/gid (e.g. node:24 has "node" at 1000:1000)
+    `existing_user=$(getent passwd ${config.ctx.uid} | cut -d: -f1) && [ -n "$existing_user" ] && [ "$existing_user" != "testuser" ] && userdel "$existing_user" 2>/dev/null || true`,
+    `existing_group=$(getent group ${config.ctx.gid} | cut -d: -f1) && [ -n "$existing_group" ] && [ "$existing_group" != "testuser" ] && groupdel "$existing_group" 2>/dev/null || true`,
     // create user matching host uid/gid for file permissions
     `id testuser > /dev/null 2>&1 || (groupadd -g ${config.ctx.gid} testuser 2>/dev/null || true; useradd -u ${config.ctx.uid} -g ${config.ctx.gid} -m -s /bin/bash testuser 2>/dev/null || true)`,
     // configure passwordless sudo (like GHA runners) - check if already configured