Thread Lifecycle

Manage Codex sessions across start, resume, fork, compact, archive, delete, and handoff.

Key takeaways

Long-running work needs thread hygiene; without lifecycle rules a session accumulates stale assumptions, duplicated plans, and unclear ownership.
Know the lifecycle actions: Start for clean context, Resume after interruption, Fork to explore a variant, Compact to reduce context load, Archive for recoverable cleanup, Delete for permanent transcript removal, and Stop to summarize final state.
In the 0.142.0-era baseline, check /usage before long goals, scheduled reminders, or broad multi-agent delegation.
In 0.145.0–0.146.0, use paginated history, persisted names/search, pins, named /new or /clear, side-conversation switching, and paginated forks to keep large thread catalogs navigable.
Make handoffs explicit with current goal and acceptance criteria, files changed, commands already run and their results, known blockers, and the next safe action.
If the thread no longer reflects actual repository state, stop and re-ground from files and git status before continuing.

Long-running work needs thread hygiene. Without lifecycle rules, the session accumulates stale assumptions, duplicated plans, and unclear ownership.

Lifecycle Actions

Action	Use
Start	New task with clean context
Resume	Continue the same task after interruption
Fork	Explore a variant without disturbing the main path
Compact	Preserve useful state while reducing context load
Archive	Remove a completed or abandoned session from active lists while keeping it recoverable
Delete	Permanently remove the transcript and spawned descendant sessions
Stop	End a task and summarize final state

/delete and codex delete <SESSION> permanently remove the transcript and spawned descendant sessions. Use /archive for recoverable cleanup.
/import can bring supported Claude Code setup, project files, and recent chats into Codex. After import, inspect /diff, /debug-config, and /status.
/usage daily|weekly|cumulative should precede long /goal sessions, scheduled reminders, and multi-agent delegation so token budgets are visible before work expands.

Experimental paginated history resumes and searches large catalogs without loading every thread, while preserving names and supporting subagent and memory metadata.
/new [name] and /clear [name] name a fresh thread immediately; pin important threads and switch side conversations without closing them.
Forks can use paginated history, and ephemeral forks stay out of normal history lists.
After replay, import, or fork, verify preservation of messages, final responses, errors, timestamps, and approval settings.

If the thread no longer reflects the actual repository state, stop and re-ground from files and git status before continuing.

Key takeaways

Long-running work needs thread hygiene; without lifecycle rules a session accumulates stale assumptions, duplicated plans, and unclear ownership.
Know the lifecycle actions: Start for clean context, Resume after interruption, Fork to explore a variant, Compact to reduce context load, Archive for recoverable cleanup, Delete for permanent transcript removal, and Stop to summarize final state.
In the 0.142.0-era baseline, check /usage before long goals, scheduled reminders, or broad multi-agent delegation.
In 0.145.0–0.146.0, use paginated history, persisted names/search, pins, named /new or /clear, side-conversation switching, and paginated forks to keep large thread catalogs navigable.
Make handoffs explicit with current goal and acceptance criteria, files changed, commands already run and their results, known blockers, and the next safe action.
If the thread no longer reflects actual repository state, stop and re-ground from files and git status before continuing.

Long-running work needs thread hygiene. Without lifecycle rules, the session accumulates stale assumptions, duplicated plans, and unclear ownership.

Action	Use
Start	New task with clean context
Resume	Continue the same task after interruption
Fork	Explore a variant without disturbing the main path
Compact	Preserve useful state while reducing context load
Archive	Remove a completed or abandoned session from active lists while keeping it recoverable
Delete	Permanently remove the transcript and spawned descendant sessions
Stop	End a task and summarize final state

/delete and codex delete <SESSION> permanently remove the transcript and spawned descendant sessions. Use /archive for recoverable cleanup.
/import can bring supported Claude Code setup, project files, and recent chats into Codex. After import, inspect /diff, /debug-config, and /status.
/usage daily|weekly|cumulative should precede long /goal sessions, scheduled reminders, and multi-agent delegation so token budgets are visible before work expands.

Experimental paginated history resumes and searches large catalogs without loading every thread, while preserving names and supporting subagent and memory metadata.
/new [name] and /clear [name] name a fresh thread immediately; pin important threads and switch side conversations without closing them.
Forks can use paginated history, and ephemeral forks stay out of normal history lists.
After replay, import, or fork, verify preservation of messages, final responses, errors, timestamps, and approval settings.

If the thread no longer reflects the actual repository state, stop and re-ground from files and git status before continuing.