Bug: codex doctor reports legacy 2025 rollout files as scan errors

Summary

codex doctor --all reports rollout scan was incomplete or found bad files for older Codex rollout JSONL files that appear to be valid legacy CLI-era conversations, but use a pre-session_meta / response_item format.

The current thread inventory is otherwise clean: active rollout files and DB rows match, archived rows match, there are no duplicate thread IDs, and all SQLite integrity checks pass.

Environment

• Codex version: 0.137.0
• Platform: macos-aarch64
• Install method: npm package @openai/codex
• Repository metadata from installed package: git+https://github.com/openai/codex.git, directory codex-cli
• Auth mode: ChatGPT auth
• Network/WebSocket/reachability: OK when run outside the local sandbox

Current Doctor Result

After quarantining unrelated context-only stubs and archived duplicates, codex doctor --all still reports:

⚠ threads      rollout scan was incomplete or found bad files
rollout DB active files  152
rollout DB active rows   152
rollout DB archived files 0
rollout DB archived rows 0
rollout DB missing active rows 0
rollout DB missing archived rows 0
rollout DB duplicate rollout thread ids 0
rollout DB duplicate DB paths 0
rollout DB scan errors   23
rollout DB malformed file names 0
rollout DB scan cap reached false

State DB health is OK:

state DB     integrity ok
log DB       integrity ok
goals DB     integrity ok
memories DB  integrity ok

Example Legacy File Shape

The scan-error files are under ~/.codex/sessions/2025/.../rollout-*.jsonl.

They start like this:

{"id":"<uuid>","timestamp":"2025-08-13T15:22:56.917Z","instructions":null}
{"record_type":"state"}
{"type":"message","id":null,"role":"user","content":[{"type":"input_text","text":"<user prompt>"}]}
{"record_type":"state"}
{"record_type":"state"}

Later records may include old-style top-level records such as:

{"type":"reasoning", ...}
{"type":"message", "role":"assistant", ...}
{"type":"function_call", ...}

These are not empty files and appear to be historical Codex CLI conversations. Newer indexed files use:

{"timestamp":"...","type":"session_meta","payload":{...}}
{"timestamp":"...","type":"response_item","payload":{...}}

Expected Behavior

One of:

• codex doctor recognizes this legacy rollout format and does not report it as a bad file.
• codex doctor reports a clearer legacy-format note instead of a warning.
• Codex provides an official migration/backfill command for legacy rollout files.
• Codex ignores legacy rollout files if they are intentionally unsupported by the current thread index.

Actual Behavior

codex doctor reports the legacy files as scan errors:

rollout DB scan error sample ~/.codex/sessions/2025/...jsonl (no parseable rollout items)

This makes the installation remain degraded even though:

• thread DB parity is otherwise clean
• active files equal active rows
• archived missing rows are zero
• duplicate thread IDs are zero
• database integrity checks pass

Why This Matters

These files appear to be historical Codex CLI conversations. Manual conversion would require rewriting JSONL record structure and inserting rows into state_5.sqlite, which is risky without an official schema/migration path.

The current doctor output suggests corruption or bad files, but the files may simply be valid legacy rollouts.

Requested Fix

Please clarify and ideally implement one of:

1. Legacy rollout support in doctor/backfill.
2. A documented migration command.
3. A clearer diagnostic such as legacy rollout format not indexed.
4. A supported way to archive/ignore legacy rollouts without doctor warnings.