omnilane 0.14.0 → 0.15.0

package/CHANGELOG.md

@@ -6,6 +6,28 @@ semantic version tags.
 
 ## [Unreleased]
 
+## [0.15.0] - 2026-08-28
+
+### Added
+
+- Timed-out Codex jobs now print the absolute path to the matching
+  `rollout-*.jsonl` session file, resolved from the first progress event's
+  `thread_id`, so an interrupted conversation remains directly inspectable.
+
+### Changed
+
+- Codex runner now invokes `codex exec --json`, streaming JSONL events to
+  `out.txt.progress.log` before the run exits or times out. `out.txt` and the
+  Jobs display, which reads only `out.txt` and `stderr.log`, remain unchanged.
+
+### Fixed
+
+- Timeout diagnostics no longer blame a usage-limit retry loop without evidence.
+  They now state that a timeout does not identify its cause, provide a
+  three-step check, and clarify that an empty progress log does not prove the
+  run made no progress.
+
+
 ## [0.14.0] - 2026-08-10
 
 ### Added
@@ -594,7 +616,9 @@ work to the wrong model, and records the evidence behind the shipped defaults.
 - Initial shared routing table, cross-vendor dispatcher, runners, installer,
   and baseline lint fixes.
 
-[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.14.0...HEAD
+[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.15.0...HEAD
+
+[0.15.0]: https://github.com/Seraphim0916/omnilane/compare/v0.14.0...v0.15.0
 
 [0.14.0]: https://github.com/Seraphim0916/omnilane/compare/v0.13.0...v0.14.0
 [0.13.0]: https://github.com/Seraphim0916/omnilane/compare/v0.12.0...v0.13.0

package/README.ja.md

@@ -516,6 +516,16 @@ scripts/dispatch.sh --dry-run hardest-coding "…"   # 解決済みプラン、
 
 ## 📜 リリース履歴
 
+## v0.15.0 の新機能
+
+- **Codex の進捗証跡をストリーミング保存** — `codex exec --json` が JSONL イベントを
+  `out.txt.progress.log` へ逐次書き込むため、タイムアウト時にも最後に到達した手順を残せます。
+  `out.txt` と Jobs の表示は変わりません。
+- **証拠に基づくタイムアウト診断** — タイムアウト自体では原因を特定できないことを明示し、
+  三段階の確認手順と、空の進捗ログが Codex の未進行を示す証拠ではないことを案内します。
+- **rollout 記録への直接パス** — 最初の進捗イベントの `thread_id` から対応する
+  `rollout-*.jsonl` の絶対パスをタイムアウト出力に表示し、中断された会話履歴を確認できます。
+
 ## v0.14.0 の新機能
 
 - **根拠に基づくルーティング提案**：`jobs recommend` と MCP `jobs_recommend` は公開ジョブメタデータだけを読み、ルーティングを自動変更しません。

package/README.ko.md

@@ -499,6 +499,16 @@ scripts/dispatch.sh --dry-run hardest-coding "…"   # 완전히 해석된 계
 
 ## 📜 릴리스 기록
 
+## v0.15.0 새 기능
+
+- **Codex 진행 증거의 스트리밍 보존** — `codex exec --json`이 JSONL 이벤트를
+  `out.txt.progress.log`에 순차 기록하므로 시간 초과가 나도 마지막으로 도달한 단계를 남깁니다.
+  `out.txt`와 Jobs 표시는 그대로입니다.
+- **증거 기반 시간 초과 진단** — 시간 초과 자체로 원인을 특정하지 않는다고 밝히고,
+  세 단계 점검 목록과 빈 진행 로그가 Codex 미진행의 증거가 아니라는 점을 안내합니다.
+- **rollout 기록으로 가는 직접 경로** — 첫 진행 이벤트의 `thread_id`로 찾은
+  `rollout-*.jsonl`의 절대 경로를 시간 초과 출력에 표시하여, 중단된 대화 이력을 확인할 수 있습니다.
+
 ## v0.14.0 새 기능
 
 - **근거 기반 라우팅 제안**: `jobs recommend`와 MCP `jobs_recommend`는 공개 작업 메타데이터만 읽고 라우팅을 자동 변경하지 않습니다.

package/README.md

@@ -205,6 +205,10 @@ header overrides that and the choice is remembered locally.
 
 Core routing does not need Python; only this UI requires Python 3.9 or newer.
 
+### Foreman completion inbox
+
+Finished dispatches write a private completion record under `$OMNILANE_HOME/inbox/`. The bundled Claude Code `UserPromptSubmit` hook atomically delivers up to ten matching records on the foreman's next prompt, scoped by the dispatch `workdir`; claimed records move to `inbox/consumed/`. Set `OMNILANE_INBOX=0` on dispatch to disable record creation. The default is enabled.
+
 ## 📦 Install
 
 Requirements: the vendor CLIs you want to route to, logged in (`codex`,
@@ -565,6 +569,18 @@ working notes, including per-benchmark caveats, live in
 
 ## 📜 Release history
 
+## What's new in v0.15.0
+
+- **Streaming Codex progress evidence** — `codex exec --json` now streams JSONL
+  events into `out.txt.progress.log`, so a timeout still records its last known
+  step. `out.txt` and the Jobs display remain unchanged.
+- **Evidence-led timeout diagnostics** — timeouts now state that they do not
+  identify the cause, list a three-step check, and clarify that an empty
+  progress log does not prove Codex made no progress.
+- **Direct rollout recovery path** — timeout output now prints the absolute path
+  to the matching `rollout-*.jsonl`, found from the first progress event's
+  `thread_id`, so the interrupted conversation history can be inspected.
+
 ## What's new in v0.14.0
 
 - **Evidence-based routing recommendations** — `jobs recommend` and MCP

package/README.zh-CN.md

@@ -473,6 +473,16 @@ vendor 一律当成 `work`,而且它只能逐次明确指定,永远不是 lane
 
 ## 📜 版本历程
 
+## v0.15.0 新功能
+
+- **流式保留 Codex 进度证据**：`codex exec --json` 会将 JSONL 事件逐条写入
+  `out.txt.progress.log`，即使超时也会留下最后执行到哪一步。`out.txt` 与任务列表显示保持不变。
+- **超时诊断回归证据**：超时提示明确说明它本身不足以判定原因，提供三步检查清单，并说明空的
+  进度日志不是 Codex 从未推进的证据。
+- **直接给出 rollout 记录位置**：超时输出会根据第一条进度事件的 `thread_id`，打印
+  ${CODEX_HOME:-$HOME/.codex}/sessions 下对应 `rollout-*.jsonl` 的绝对路径，便于查看
+  被中断而未回报的完整对话历程。
+
 ## v0.14.0 新功能
 
 - **基于证据的路由建议**：`jobs recommend` 与 MCP `jobs_recommend` 只读取公开作业元数据，并且不会自动修改路由。

package/README.zh-TW.md

@@ -497,6 +497,16 @@ vendor 一律當成 `work`,而且它只能逐次明確指定,永遠不是 lane
 
 ## 📜 版本歷程
 
+## v0.15.0 新功能
+
+- **串流保留 Codex 進度證據**：`codex exec --json` 會把 JSONL 事件逐筆寫入
+  `out.txt.progress.log`，即使逾時也能留下最後做到哪一步。`out.txt` 與工作清單顯示維持不變。
+- **逾時診斷回到證據**：逾時訊息明示它本身不足以判定原因，提供三步檢查清單，並說明空的
+  進度日誌不是 Codex 未曾前進的證據。
+- **直接指出 rollout 記錄位置**：逾時輸出會依第一筆進度事件的 `thread_id`，印出
+  ${CODEX_HOME:-$HOME/.codex}/sessions 下對應 `rollout-*.jsonl` 的絕對路徑，方便查看
+  被中斷而未回報的完整對話歷程。
+
 ## v0.14.0 新功能
 
 - **依證據產生路由建議**：`jobs recommend` 與 MCP `jobs_recommend` 只用已完成工作的

package/VERSION

@@ -1 +1 @@
-0.14.0
+0.15.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnilane",
-  "version": "0.14.0",
+  "version": "0.15.0",
   "description": "One routing table, every harness — classify subtasks into lanes and dispatch each lane to the best vendor's agentic CLI (Codex, Claude, Gemini, Grok) using your existing subscription logins.",
   "bin": {
     "omnilane": "bin/omnilane"

package/scripts/dispatch.sh

@@ -655,10 +655,58 @@ secure_job_files() {
   find "$JOB_DIR" -type f -exec chmod 600 {} +
 }
 
+completion_tail() {
+  local path="$1" size=0 keep value="" sentinel=$'\001'
+  local note=$'[truncated: leading output omitted]\n'
+  local LC_ALL=C
+
+  if [[ -f "$path" && ! -L "$path" ]]; then
+    size="$(wc -c < "$path" 2>/dev/null | tr -d '[:space:]')" || return 1
+    [[ "$size" =~ ^[0-9]+$ ]] || return 1
+    if [[ "$size" -gt 2000 ]]; then
+      keep=$((2000 - ${#note}))
+      value="$(printf '%s' "$note"; tail -c "$keep" "$path"; printf '%s' "$sentinel")" || return 1
+    else
+      value="$(cat "$path"; printf '%s' "$sentinel")" || return 1
+    fi
+    value="${value%"$sentinel"}"
+  fi
+  printf '%s' "$value"
+}
+
+write_completion_record() {
+  local rc="$1" inbox="$OMNILANE_HOME/inbox"
+  local final="$inbox/$JOB_ID.json" tmp="$inbox/.$JOB_ID.tmp.$$-$RANDOM"
+  local tail_value finished old_umask write_rc=0 sentinel=$'\001'
+
+  prepare_inbox_store || return 1
+  tail_value="$(completion_tail "$JOB_DIR/out.txt"; printf '%s' "$sentinel")" || return 1
+  tail_value="${tail_value%"$sentinel"}"
+  finished="$(date -u +%FT%TZ)" || return 1
+  old_umask="$(umask)"
+  umask 077
+  printf '{"job_id":"%s","lane":"%s","vendor":"%s","model":"%s","mode":"%s","workdir":"%s","exit":%s,"finished":"%s","tail":"%s"}\n' \
+    "$(json_escape "$JOB_ID")" "$(json_escape "$LANE")" "$(json_escape "$VENDOR")" \
+    "$(json_escape "$MODEL")" "$(json_escape "$MODE")" "$(json_escape "$WORKDIR")" \
+    "$rc" "$(json_escape "$finished")" "$(json_escape "$tail_value")" > "$tmp" || write_rc=$?
+  if [[ "$write_rc" -eq 0 ]]; then
+    chmod 600 "$tmp" || write_rc=$?
+  fi
+  if [[ "$write_rc" -eq 0 ]]; then
+    mv "$tmp" "$final" || write_rc=$?
+  fi
+  umask "$old_umask"
+  rm -f "$tmp" 2>/dev/null || true
+  return "$write_rc"
+}
+
 finish_job() {
   local rc="$1"
   secure_job_files
   (umask 077; printf '%s\n' "$rc" > "$JOB_DIR/exit")
+  if [[ "${OMNILANE_INBOX:-1}" != "0" ]]; then
+    write_completion_record "$rc" >/dev/null 2>&1 || true
+  fi
 }
 
 run_job() {

package/scripts/lib/common.sh

@@ -181,6 +181,25 @@ read_lock_owner() {
   LOCK_OWNER_VALUE="$value"
 }
 
+prepare_private_store() { # path, diagnostic label
+  local store_root="$1" store_label="$2"
+  mkdir -p "$OMNILANE_HOME" || return 1
+  if [[ -L "$store_root" || ( -e "$store_root" && ! -d "$store_root" ) ]]; then
+    echo "omnilane: unsafe $store_label path (want real directory): $store_root" >&2
+    return 1
+  fi
+  [[ -d "$store_root" ]] || mkdir -m 700 "$store_root" || return 1
+  [[ -d "$store_root" && ! -L "$store_root" ]] || {
+    echo "omnilane: $store_label changed while preparing it" >&2
+    return 1
+  }
+  chmod 700 "$store_root" || return 1
+}
+
+prepare_inbox_store() {
+  prepare_private_store "$OMNILANE_HOME/inbox" "inbox store"
+}
+
 prepare_lock_store() {
   local lock_root="$OMNILANE_HOME/locks"
   mkdir -p "$OMNILANE_HOME"

package/scripts/runners/run-codex.sh

@@ -16,7 +16,9 @@ RUN_TIMEOUT="${OMNILANE_TIMEOUT:-600}"
 
 # --skip-git-repo-check: the operator chose WORKDIR explicitly; codex would
 # otherwise refuse any directory that is not a trusted git repo.
-ARGS=(exec -m "$MODEL" -o "${OUTPUT_FILE}.tmp" --skip-git-repo-check)
+# --json: without it codex writes nothing until it exits, so a watchdog kill
+# leaves an empty progress log that looks identical to a run that never started.
+ARGS=(exec --json -m "$MODEL" -o "${OUTPUT_FILE}.tmp" --skip-git-repo-check)
 [[ -n "$EFFORT" && "$EFFORT" != "-" ]] && ARGS+=(-c "model_reasoning_effort=\"$EFFORT\"")
 if [[ "$MODE" == "advise" ]]; then
   ARGS+=(--ephemeral -s read-only)
@@ -41,10 +43,25 @@ set +e
 RC=$?
 set -e
 
-# Watchdog kill with an empty progress log = codex produced nothing at all;
-# the usual cause is a quota-window 429 retry loop, which codex retries silently.
-if [[ ("$RC" -eq 142 || "$RC" -eq 124) && ! -s "${OUTPUT_FILE}.progress.log" ]]; then
-  echo "omnilane: codex timed out after ${RUN_TIMEOUT}s with no output — likely a usage-limit retry loop; check your codex quota window and retry" >> "${OUTPUT_FILE}.stderr.log"
+if [[ "$RC" -eq 142 || "$RC" -eq 124 ]]; then
+  {
+    echo "omnilane: codex timed out after ${RUN_TIMEOUT}s. This does NOT identify the cause."
+    echo "omnilane: an empty progress log is NOT evidence of a stall — check, in order:"
+    echo "  1. the streamed events in ${OUTPUT_FILE}.progress.log — the last event says how far the run got"
+    echo "  2. the rollout file below — if it stops at task_started, the request never left the CLI"
+    echo "  3. your provider proxy's log for the run window: rate-limit responses, and whether a"
+    echo "     request was ever sent upstream at all"
+    # The rollout holds the full turn history a killed run never got to report;
+    # thread_id is the only handle onto it once the process is gone.
+    thread_id="$(head -1 "${OUTPUT_FILE}.progress.log" 2>/dev/null |
+      sed -n 's/.*"thread_id":"\([^"]*\)".*/\1/p')"
+    if [[ -n "$thread_id" ]]; then
+      rollout="$(find "${CODEX_HOME:-$HOME/.codex}/sessions" -name "rollout-*-${thread_id}.jsonl" 2>/dev/null | head -1)"
+      echo "omnilane: rollout: ${rollout:-not found (thread ${thread_id})}"
+    else
+      echo "omnilane: rollout: unknown — no thread_id in the progress log, so codex died before its first event"
+    fi
+  } >> "${OUTPUT_FILE}.stderr.log"
 fi
 
 if [[ -f "${OUTPUT_FILE}.tmp" ]]; then