Make AGENTS.md discovery FS-aware (#15826)

## Summary
- make AGENTS.md discovery and loading fully FS-aware and remove the
non-FS discover helper
- migrate remote-aware codex-core tests to use TestEnv workspace setup
instead of syncing a local workspace copy
- add AGENTS.md corner-case coverage, including directory fallbacks and
remote-aware integration coverage

## Testing
- cargo test -p codex-core project_doc -- --nocapture
- cargo test -p codex-core hierarchical_agents -- --nocapture
- cargo test -p codex-core agents_md -- --nocapture
- cargo test -p codex-tui status -- --nocapture
- cargo test -p codex-tui-app-server status -- --nocapture
- just fix
- just fmt
- just bazel-lock-update
- just bazel-lock-check
- just argument-comment-lint
- remote Linux executor tests in progress via scripts/test-remote-env.sh
This commit is contained in:
pakrym-oai
2026-04-06 20:26:21 -07:00
committed by GitHub
parent 232db0613a
commit 4bb507d2c4
21 changed files with 545 additions and 171 deletions
+74 -44
View File
@@ -1,3 +1,4 @@
use std::future::Future;
use std::mem::swap;
use std::path::Path;
use std::path::PathBuf;
@@ -34,12 +35,12 @@ use codex_protocol::protocol::SessionConfiguredEvent;
use codex_protocol::protocol::SessionSource;
use codex_protocol::user_input::UserInput;
use codex_utils_absolute_path::AbsolutePathBuf;
use futures::future::BoxFuture;
use serde_json::Value;
use tempfile::TempDir;
use wiremock::MockServer;
use crate::PathBufExt;
use crate::PathExt;
use crate::RemoteEnvConfig;
use crate::TempDirExt;
use crate::get_remote_test_env;
@@ -55,6 +56,8 @@ use wiremock::matchers::path_regex;
type ConfigMutator = dyn FnOnce(&mut Config) + Send;
type PreBuildHook = dyn FnOnce(&Path) + Send + 'static;
type WorkspaceSetup = dyn FnOnce(AbsolutePathBuf, Arc<dyn ExecutorFileSystem>) -> BoxFuture<'static, Result<()>>
+ Send;
const TEST_MODEL_WITH_EXPERIMENTAL_TOOLS: &str = "test-gpt-5.1-codex";
const REMOTE_EXEC_SERVER_START_TIMEOUT: Duration = Duration::from_secs(5);
const REMOTE_EXEC_SERVER_POLL_INTERVAL: Duration = Duration::from_millis(25);
@@ -97,24 +100,28 @@ impl RemoteExecServerProcess {
#[derive(Debug)]
pub struct TestEnv {
environment: codex_exec_server::Environment,
cwd: PathBuf,
_local_cwd_temp_dir: Option<TempDir>,
cwd: AbsolutePathBuf,
local_cwd_temp_dir: Option<Arc<TempDir>>,
_remote_exec_server_process: Option<RemoteExecServerProcess>,
}
impl TestEnv {
pub async fn local() -> Result<Self> {
let local_cwd_temp_dir = TempDir::new()?;
let cwd = local_cwd_temp_dir.path().to_path_buf();
let local_cwd_temp_dir = Arc::new(TempDir::new()?);
let cwd = local_cwd_temp_dir.abs();
let environment = codex_exec_server::Environment::create(/*exec_server_url*/ None).await?;
Ok(Self {
environment,
cwd,
_local_cwd_temp_dir: Some(local_cwd_temp_dir),
local_cwd_temp_dir: Some(local_cwd_temp_dir),
_remote_exec_server_process: None,
})
}
pub fn cwd(&self) -> &AbsolutePathBuf {
&self.cwd
}
pub fn environment(&self) -> &codex_exec_server::Environment {
&self.environment
}
@@ -122,6 +129,10 @@ impl TestEnv {
pub fn exec_server_url(&self) -> Option<&str> {
self.environment.exec_server_url()
}
fn local_cwd_temp_dir(&self) -> Option<Arc<TempDir>> {
self.local_cwd_temp_dir.clone()
}
}
pub async fn test_env() -> Result<TestEnv> {
@@ -134,16 +145,13 @@ pub async fn test_env() -> Result<TestEnv> {
let cwd = remote_aware_cwd_path();
environment
.get_filesystem()
.create_directory(
&absolute_path(&cwd)?,
CreateDirectoryOptions { recursive: true },
)
.create_directory(&cwd, CreateDirectoryOptions { recursive: true })
.await?;
remote_process.process.register_cleanup_path(&cwd);
remote_process.process.register_cleanup_path(cwd.as_path());
Ok(TestEnv {
environment,
cwd,
_local_cwd_temp_dir: None,
local_cwd_temp_dir: None,
_remote_exec_server_process: Some(remote_process.process),
})
}
@@ -201,11 +209,12 @@ echo $!"
})
}
fn remote_aware_cwd_path() -> PathBuf {
fn remote_aware_cwd_path() -> AbsolutePathBuf {
PathBuf::from(format!(
"/tmp/codex-core-test-cwd-{}",
remote_exec_server_instance_id()
))
.abs()
}
fn wait_for_remote_listen_url(container_name: &str, stdout_path: &str) -> Result<String> {
@@ -299,10 +308,6 @@ fn docker_command_capture_stdout<const N: usize>(args: [&str; N]) -> Result<Stri
String::from_utf8(output.stdout).context("docker stdout must be utf-8")
}
fn absolute_path(path: &Path) -> Result<AbsolutePathBuf> {
Ok(path.abs())
}
/// A collection of different ways the model can output an apply_patch call
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub enum ApplyPatchModelOutput {
@@ -326,6 +331,7 @@ pub struct TestCodexBuilder {
config_mutators: Vec<Box<ConfigMutator>>,
auth: CodexAuth,
pre_build_hooks: Vec<Box<PreBuildHook>>,
workspace_setups: Vec<Box<WorkspaceSetup>>,
home: Option<Arc<TempDir>>,
user_shell_override: Option<Shell>,
}
@@ -359,6 +365,16 @@ impl TestCodexBuilder {
self
}
pub fn with_workspace_setup<F, Fut>(mut self, setup: F) -> Self
where
F: FnOnce(AbsolutePathBuf, Arc<dyn ExecutorFileSystem>) -> Fut + Send + 'static,
Fut: Future<Output = Result<()>> + Send + 'static,
{
self.workspace_setups
.push(Box::new(move |cwd, fs| Box::pin(setup(cwd, fs))));
self
}
pub fn with_home(mut self, home: Arc<TempDir>) -> Self {
self.home = Some(home);
self
@@ -382,25 +398,24 @@ impl TestCodexBuilder {
Some(home) => home,
None => Arc::new(TempDir::new()?),
};
Box::pin(self.build_with_home(server, home, /*resume_from*/ None)).await
let base_url = format!("{}/v1", server.uri());
let test_env = TestEnv::local().await?;
Box::pin(self.build_with_home_and_base_url(base_url, home, /*resume_from*/ None, test_env))
.await
}
pub async fn build_remote_aware(
&mut self,
server: &wiremock::MockServer,
) -> anyhow::Result<TestCodex> {
let test_env = test_env().await?;
let home = match self.home.clone() {
Some(home) => home,
None => Arc::new(TempDir::new()?),
};
let base_url = format!("{}/v1", server.uri());
let cwd = test_env.cwd.to_path_buf();
self.config_mutators.push(Box::new(move |config| {
config.cwd = cwd.abs();
}));
let (config, cwd) = self.prepare_config(base_url, &home).await?;
Box::pin(self.build_from_config(config, cwd, home, /*resume_from*/ None, test_env)).await
let test_env = test_env().await?;
Box::pin(self.build_with_home_and_base_url(base_url, home, /*resume_from*/ None, test_env))
.await
}
pub async fn build_with_streaming_server(
@@ -412,10 +427,12 @@ impl TestCodexBuilder {
Some(home) => home,
None => Arc::new(TempDir::new()?),
};
let test_env = TestEnv::local().await?;
Box::pin(self.build_with_home_and_base_url(
format!("{base_url}/v1"),
home,
/*resume_from*/ None,
test_env,
))
.await
}
@@ -435,7 +452,9 @@ impl TestCodexBuilder {
config.model_provider.supports_websockets = true;
config.experimental_realtime_ws_model = Some("realtime-test-model".to_string());
}));
Box::pin(self.build_with_home_and_base_url(base_url, home, /*resume_from*/ None)).await
let test_env = TestEnv::local().await?;
Box::pin(self.build_with_home_and_base_url(base_url, home, /*resume_from*/ None, test_env))
.await
}
pub async fn resume(
@@ -443,19 +462,10 @@ impl TestCodexBuilder {
server: &wiremock::MockServer,
home: Arc<TempDir>,
rollout_path: PathBuf,
) -> anyhow::Result<TestCodex> {
Box::pin(self.build_with_home(server, home, Some(rollout_path))).await
}
async fn build_with_home(
&mut self,
server: &wiremock::MockServer,
home: Arc<TempDir>,
resume_from: Option<PathBuf>,
) -> anyhow::Result<TestCodex> {
let base_url = format!("{}/v1", server.uri());
let (config, cwd) = self.prepare_config(base_url, &home).await?;
Box::pin(self.build_from_config(config, cwd, home, resume_from, TestEnv::local().await?))
let test_env = TestEnv::local().await?;
Box::pin(self.build_with_home_and_base_url(base_url, home, Some(rollout_path), test_env))
.await
}
@@ -464,10 +474,30 @@ impl TestCodexBuilder {
base_url: String,
home: Arc<TempDir>,
resume_from: Option<PathBuf>,
test_env: TestEnv,
) -> anyhow::Result<TestCodex> {
let (config, cwd) = self.prepare_config(base_url, &home).await?;
Box::pin(self.build_from_config(config, cwd, home, resume_from, TestEnv::local().await?))
.await
let (config, fallback_cwd) = self
.prepare_config(base_url, &home, test_env.cwd().clone())
.await?;
let environment_manager = Arc::new(codex_exec_server::EnvironmentManager::new(
test_env.exec_server_url().map(str::to_owned),
));
let file_system = test_env.environment().get_filesystem();
let mut workspace_setups = vec![];
swap(&mut self.workspace_setups, &mut workspace_setups);
for setup in workspace_setups {
setup(config.cwd.clone(), Arc::clone(&file_system)).await?;
}
let cwd = test_env.local_cwd_temp_dir().unwrap_or(fallback_cwd);
Box::pin(self.build_from_config(
config,
cwd,
home,
resume_from,
test_env,
environment_manager,
))
.await
}
async fn build_from_config(
@@ -477,11 +507,9 @@ impl TestCodexBuilder {
home: Arc<TempDir>,
resume_from: Option<PathBuf>,
test_env: TestEnv,
environment_manager: Arc<codex_exec_server::EnvironmentManager>,
) -> anyhow::Result<TestCodex> {
let auth = self.auth.clone();
let environment_manager = Arc::new(codex_exec_server::EnvironmentManager::new(
test_env.exec_server_url().map(str::to_owned),
));
let thread_manager = if config.model_catalog.is_some() {
ThreadManager::new(
&config,
@@ -553,6 +581,7 @@ impl TestCodexBuilder {
&mut self,
base_url: String,
home: &TempDir,
cwd_override: AbsolutePathBuf,
) -> anyhow::Result<(Config, Arc<TempDir>)> {
let model_provider = ModelProviderInfo {
base_url: Some(base_url),
@@ -563,7 +592,7 @@ impl TestCodexBuilder {
};
let cwd = Arc::new(TempDir::new()?);
let mut config = load_default_config_for_test(home).await;
config.cwd = cwd.abs();
config.cwd = cwd_override;
config.model_provider = model_provider;
for hook in self.pre_build_hooks.drain(..) {
hook(home.path());
@@ -897,6 +926,7 @@ pub fn test_codex() -> TestCodexBuilder {
config_mutators: vec![],
auth: CodexAuth::from_api_key("dummy"),
pre_build_hooks: vec![],
workspace_setups: vec![],
home: None,
user_shell_override: None,
}
@@ -17,6 +17,7 @@ impl TestCodexExecBuilder {
);
cmd.current_dir(self.cwd.path())
.env("CODEX_HOME", self.home.path())
.env("CODEX_SQLITE_HOME", self.home.path())
.env(CODEX_API_KEY_ENV_VAR, "dummy");
cmd
}
+130
View File
@@ -0,0 +1,130 @@
use anyhow::Result;
use codex_exec_server::CreateDirectoryOptions;
use core_test_support::responses::ev_completed;
use core_test_support::responses::ev_response_created;
use core_test_support::responses::mount_sse_once;
use core_test_support::responses::sse;
use core_test_support::responses::start_mock_server;
use core_test_support::test_codex::TestCodexBuilder;
use core_test_support::test_codex::test_codex;
async fn agents_instructions(mut builder: TestCodexBuilder) -> Result<String> {
let server = start_mock_server().await;
let resp_mock = mount_sse_once(
&server,
sse(vec![ev_response_created("resp1"), ev_completed("resp1")]),
)
.await;
let test = builder.build_remote_aware(&server).await?;
test.submit_turn("hello").await?;
let request = resp_mock.single_request();
request
.message_input_texts("user")
.into_iter()
.find(|text| text.starts_with("# AGENTS.md instructions for "))
.ok_or_else(|| anyhow::anyhow!("instructions message not found"))
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn agents_override_is_preferred_over_agents_md() -> Result<()> {
let instructions =
agents_instructions(test_codex().with_workspace_setup(|cwd, fs| async move {
let agents_md = cwd.join("AGENTS.md").expect("absolute AGENTS.md path");
let override_md = cwd
.join("AGENTS.override.md")
.expect("absolute AGENTS.override.md path");
fs.write_file(&agents_md, b"base doc".to_vec()).await?;
fs.write_file(&override_md, b"override doc".to_vec())
.await?;
Ok::<(), anyhow::Error>(())
}))
.await?;
assert!(
instructions.contains("override doc"),
"expected AGENTS.override.md contents: {instructions}"
);
assert!(
!instructions.contains("base doc"),
"expected AGENTS.md to be ignored when override exists: {instructions}"
);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn configured_fallback_is_used_when_agents_candidate_is_directory() -> Result<()> {
let instructions = agents_instructions(
test_codex()
.with_config(|config| {
config.project_doc_fallback_filenames = vec!["WORKFLOW.md".to_string()];
})
.with_workspace_setup(|cwd, fs| async move {
let agents_dir = cwd.join("AGENTS.md").expect("absolute AGENTS.md path");
let fallback = cwd.join("WORKFLOW.md").expect("absolute WORKFLOW.md path");
fs.create_directory(&agents_dir, CreateDirectoryOptions { recursive: true })
.await?;
fs.write_file(&fallback, b"fallback doc".to_vec()).await?;
Ok::<(), anyhow::Error>(())
}),
)
.await?;
assert!(
instructions.contains("fallback doc"),
"expected fallback doc contents: {instructions}"
);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn agents_docs_are_concatenated_from_project_root_to_cwd() -> Result<()> {
let instructions = agents_instructions(
test_codex()
.with_config(|config| {
config.cwd = config
.cwd
.join("nested/workspace")
.expect("absolute nested workspace path");
})
.with_workspace_setup(|cwd, fs| async move {
let nested = cwd.clone();
let root = nested
.parent()
.and_then(|parent| parent.parent())
.expect("nested workspace should have a project root ancestor");
let root_agents = root
.join("AGENTS.md")
.expect("absolute root AGENTS.md path");
let git_marker = root.join(".git").expect("absolute .git path");
let nested_agents = nested
.join("AGENTS.md")
.expect("absolute nested AGENTS.md path");
fs.create_directory(&nested, CreateDirectoryOptions { recursive: true })
.await?;
fs.write_file(&root_agents, b"root doc".to_vec()).await?;
fs.write_file(&git_marker, b"gitdir: /tmp/mock-git-dir\n".to_vec())
.await?;
fs.write_file(&nested_agents, b"child doc".to_vec()).await?;
Ok::<(), anyhow::Error>(())
}),
)
.await?;
let root_pos = instructions
.find("root doc")
.expect("expected root doc in AGENTS instructions");
let child_pos = instructions
.find("child doc")
.expect("expected child doc in AGENTS instructions");
assert!(
root_pos < child_pos,
"expected root doc before child doc: {instructions}"
);
Ok(())
}
@@ -18,21 +18,22 @@ async fn hierarchical_agents_appends_to_project_doc_in_user_instructions() {
)
.await;
let mut builder = test_codex().with_config(|config| {
config
.features
.enable(Feature::ChildAgentsMd)
.expect("test config should allow feature update");
std::fs::write(
let mut builder = test_codex()
.with_config(|config| {
config
.cwd
.join("AGENTS.md")
.expect("absolute AGENTS.md path"),
"be nice",
)
.expect("write AGENTS.md");
});
let test = builder.build(&server).await.expect("build test codex");
.features
.enable(Feature::ChildAgentsMd)
.expect("test config should allow feature update");
})
.with_workspace_setup(|cwd, fs| async move {
let agents_md = cwd.join("AGENTS.md").expect("absolute AGENTS.md path");
fs.write_file(&agents_md, b"be nice".to_vec()).await?;
Ok::<(), anyhow::Error>(())
});
let test = builder
.build_remote_aware(&server)
.await
.expect("build test codex");
test.submit_turn("hello").await.expect("submit turn");
@@ -73,7 +74,10 @@ async fn hierarchical_agents_emits_when_no_project_doc() {
.enable(Feature::ChildAgentsMd)
.expect("test config should allow feature update");
});
let test = builder.build(&server).await.expect("build test codex");
let test = builder
.build_remote_aware(&server)
.await
.expect("build test codex");
test.submit_turn("hello").await.expect("submit turn");
+1
View File
@@ -74,6 +74,7 @@ pub static CODEX_ALIASES_TEMP_DIR: Option<TestCodexAliasesGuard> = {
mod abort_tasks;
mod agent_jobs;
mod agent_websocket;
mod agents_md;
mod apply_patch_cli;
#[cfg(not(target_os = "windows"))]
mod approvals;
+2 -9
View File
@@ -1,6 +1,6 @@
use anyhow::Result;
use codex_exec_server::RemoveOptions;
use codex_utils_absolute_path::AbsolutePathBuf;
use core_test_support::PathBufExt;
use core_test_support::get_remote_test_env;
use core_test_support::test_codex::test_env;
use pretty_assertions::assert_eq;
@@ -17,8 +17,7 @@ async fn remote_test_env_can_connect_and_use_filesystem() -> Result<()> {
let test_env = test_env().await?;
let file_system = test_env.environment().get_filesystem();
let file_path = remote_test_file_path();
let file_path_abs = absolute_path(file_path.clone())?;
let file_path_abs = remote_test_file_path().abs();
let payload = b"remote-test-env-ok".to_vec();
file_system
@@ -39,12 +38,6 @@ async fn remote_test_env_can_connect_and_use_filesystem() -> Result<()> {
Ok(())
}
fn absolute_path(path: PathBuf) -> Result<AbsolutePathBuf> {
AbsolutePathBuf::try_from(path.clone())
.map_err(|err| anyhow::anyhow!("invalid absolute path {}: {err}", path.display()))
}
fn remote_test_file_path() -> PathBuf {
let nanos = match SystemTime::now().duration_since(UNIX_EPOCH) {
Ok(duration) => duration.as_nanos(),