Codex Memories stage1 succeeds but global consolidation remains pending and no summary artifact is generated on Windows

Summary

Codex Memories appears to be enabled and stage1 extraction is working, but global consolidation does not produce a summary artifact.

This may be related to #21884, but my local state is slightly different: stage1_outputs is not empty and raw memory files are being generated. The issue appears to be in the global consolidation / summary artifact step.

Environment

• OS: Windows
• Codex CLI: 0.130.0
• Auth mode: ChatGPT
• Memories feature: enabled

Config:

[features]
memories = true

[memories]
generate_memories = true
use_memories = true

Observed behavior

The following files/directories exist and are populated:

• ~/.codex/memories/raw_memories.md
• ~/.codex/memories/rollout_summaries/

The local state database also shows memory extraction output:

• stage1_outputs: 24 rows
• multiple memory_stage1 jobs are done

However, the following summary artifacts are not present:

• ~/.codex/memories/MEMORY.md
• ~/.codex/memories/memory_summary.md

The jobs table shows:

• memory_consolidate_global with job_key = global
• status: pending
• last_success_watermark appears to be set
• no generated summary artifact appears in ~/.codex/memories/

Some memory_stage1 jobs failed with:

Codex ran out of room in the model's context window. Start a new thread or clear earlier history before retrying.

Expected behavior

If stage1 memory extraction succeeds, Codex should either:

1. generate/update the summary artifact used by future sessions, or
2. surface a clear error explaining why global consolidation cannot complete.

Actual behavior

Raw memory files and rollout summaries are created, but the global summary artifact does not appear to be generated. The settings indicate Memories are enabled, but it is unclear whether future sessions can actually use the extracted memories.

Notes

This looks different from a complete memory extraction failure because stage1_outputs contains rows and raw_memories.md is populated. The problem seems to be at or after memory_consolidate_global.