Skip to content

armgit / Wake

Public Imported from GitHub
0

All your AI agent sessions in one place — browse, search, resume. Rust + GPUI.

Rust

fix(codex): recognize the usage and voice records Codex 0.156 writes

Corey Chiu733832253e3c
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 00f4cd4..9f3249b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,7 @@
## [Unreleased]
+- Fix: Clean Up no longer refuses Codex sessions written by recent versions
- Fix: Open In → iTerm resumes the session in a new iTerm window instead of failing with an AppleScript error (#73, thanks @quintus-lab)
- Fix: Conversations from the Antigravity CLI show their full messages, not just a summary (#71, thanks @dingtwo)
diff --git a/crates/wake-core/src/adapters/codex.rs b/crates/wake-core/src/adapters/codex.rs
index 42ba282..de2782a 100644
--- a/crates/wake-core/src/adapters/codex.rs
+++ b/crates/wake-core/src/adapters/codex.rs
@@ -540,6 +540,8 @@ fn parse_rollout(path: &Path, decode_images: bool) -> Result<CodexParse> {
let mut history_start: Option<i64> = None;
// 子线程里"父线程历史到此为止"的下标(messages / event_fallback 各一)
let mut inherited: Option<(usize, usize)> = None;
+ // 用量记录里整个线程的累计,只在文件里没有 token_count 时用
+ let mut usage_record_tokens: i64 = 0;
for row in jsonl_values(reader) {
let Some(row) = row? else {
@@ -875,9 +877,30 @@ fn parse_rollout(path: &Path, decode_images: bool) -> Result<CodexParse> {
}
// 子线程每收一次派活就有一行,内容只有 {"trigger_turn":true}
"world_state" | "inter_agent_communication_metadata" => {}
+ // 0.156 起每次模型调用记一行用量(这一次 / 这一轮 / 整个线程三份)。token 仍以
+ // event_msg 的 token_count 为准:两种在同一文件里同时出现(本机 51 份全是),整个线程
+ // 那份又与 token_count 不总相等(14 份多出约 2%)。一份文件只认一个来源,这里只记下
+ // 来,哪天 Codex 不写 token_count 了才用它
+ "token_usage_record" => {
+ if let Some(total) = payload
+ .pointer("/thread_token_usage/total_tokens")
+ .and_then(Value::as_i64)
+ {
+ usage_record_tokens = total;
+ }
+ }
+ // 语音会话的起止标记;transcript_segment 这类带内容的子类型还没见过样本,照旧计未知行
+ "realtime_item"
+ if matches!(
+ payload.get("type").and_then(Value::as_str),
+ Some("realtime_session_started" | "realtime_session_closed")
+ ) => {}
_ => unknown_lines += 1,
}
}
+ if tokens_used == 0 {
+ tokens_used = usage_record_tokens;
+ }
// fork 段先折叠再判 has_real:折完只剩一条 Meta 的子线程要走得到
// event_msg 回退,与普通会话同一条路。回退流只截断、**不留标记**——
diff --git a/crates/wake-core/tests/adapter_contracts.rs b/crates/wake-core/tests/adapter_contracts.rs
index 0f101fd..bf40563 100644
--- a/crates/wake-core/tests/adapter_contracts.rs
+++ b/crates/wake-core/tests/adapter_contracts.rs
@@ -1483,6 +1483,87 @@ fn codex_review_output_is_readable_and_not_duplicated() {
assert!(session.units[0].text.contains("Keep the selected row"));
}
+/// 0.156 起的两种记录:`token_usage_record` 每次模型调用一行,`realtime_item` 的起止
+/// 标记是语音会话的元数据,都不算未知行(Clean Up 移动前要求整份读懂,否则拒清)。
+/// token 以 event_msg 的 token_count 为准,用量记录里整个线程的累计只在文件里没有
+/// token_count 时才用;带内容的语音子类型(transcript_segment 等)还没见过样本,照旧计未知行
+#[test]
+fn codex_usage_and_realtime_records_are_known() {
+ setup();
+ let home = tempfile::tempdir().unwrap();
+ let day = home.path().join("sessions/2026/10/19");
+ fs::create_dir_all(&day).unwrap();
+ let id = "01a12345-0000-7000-8000-000000000156";
+ let usage = |input: i64| {
+ serde_json::json!({
+ "input_tokens": input, "cached_input_tokens": 0, "cache_write_input_tokens": 0,
+ "output_tokens": 10, "reasoning_output_tokens": 0, "total_tokens": input + 10
+ })
+ };
+ let realtime = |kind: &str| {
+ serde_json::json!({"timestamp": "2026-10-19T00:00:05.000Z", "type": "realtime_item",
+ "payload": {"realtime_session_id": "rs-1", "type": kind}})
+ };
+ let token_count = serde_json::json!({"timestamp": "2026-10-19T00:00:04.000Z", "type": "event_msg",
+ "payload": {"type": "token_count", "info": {"total_token_usage": usage(3000),
+ "last_token_usage": usage(100), "model_context_window": 272000}}});
+ let lines = vec![
+ serde_json::json!({"timestamp": "2026-10-19T00:00:00.000Z", "type": "session_meta",
+ "payload": {"id": id, "timestamp": "2026-10-19T00:00:00.000Z", "cwd": "/work/wake",
+ "originator": "codex_cli_rs", "source": "cli", "cli_version": "0.156.0"}}),
+ realtime("realtime_session_started"),
+ serde_json::json!({"timestamp": "2026-10-19T00:00:02.000Z", "type": "response_item",
+ "payload": {"type": "message", "role": "user",
+ "content": [{"type": "input_text", "text": "Count the tokens"}]}}),
+ serde_json::json!({"timestamp": "2026-10-19T00:00:03.000Z", "type": "response_item",
+ "payload": {"type": "message", "role": "assistant",
+ "content": [{"type": "output_text", "text": "Done."}]}}),
+ serde_json::json!({"timestamp": "2026-10-19T00:00:03.500Z", "type": "token_usage_record",
+ "payload": {"thread_id": id, "turn_id": "t-1", "session_id": id, "root_turn_id": "t-1",
+ "response_id": "resp_1", "usage": usage(100), "turn_token_usage": usage(100),
+ "thread_token_usage": usage(4000)}}),
+ token_count.clone(),
+ realtime("realtime_session_closed"),
+ realtime("transcript_segment"),
+ ];
+ let path = day.join(format!("rollout-2026-10-19T00-00-00-{id}.jsonl"));
+ let write = |lines: &[serde_json::Value]| {
+ let body: String = lines.iter().map(|l| format!("{l}\n")).collect();
+ fs::write(&path, body).unwrap();
+ };
+ write(&lines);
+
+ let adapter = CodexAdapter::new().with_custom_root(home.path().to_path_buf());
+ let r = adapter.file_ref(&path).expect("用户线程照常列出");
+ let transcript = adapter.parse_transcript(&r).unwrap();
+ assert_eq!(
+ transcript.unknown_line_count, 1,
+ "只有 transcript_segment 计未知行"
+ );
+ let texts: Vec<&str> = transcript
+ .mainline
+ .iter()
+ .filter(|m| m.kind != MessageKind::Meta)
+ .map(|m| m.text.as_str())
+ .collect();
+ assert_eq!(texts, ["Count the tokens", "Done."]);
+ let session = adapter.parse_session(&r).unwrap();
+ assert_eq!(
+ session.meta.tokens_used,
+ Some(3010),
+ "token 以 token_count 的累计为准"
+ );
+
+ // 哪天 Codex 不写 token_count 了:退回用量记录里整个线程的累计,不至于没有 token
+ let without: Vec<serde_json::Value> = lines.into_iter().filter(|l| *l != token_count).collect();
+ write(&without);
+ let r = adapter.file_ref(&path).unwrap();
+ assert_eq!(
+ adapter.parse_session(&r).unwrap().meta.tokens_used,
+ Some(4010)
+ );
+}
+
#[test]
fn qoder_parse_contract() {
setup();