Continue Codex Work

Overview

Recover actionable context from a prior Codex CLI session and continue execution in the current conversation. Codex records each session as a rollout JSONL under ~/.codex/sessions/YYYY/MM/DD/rollout-*.jsonl (with an optional state_*.sqlite index). Use those local files as the source of truth, then continue with concrete edits and checks — not just summarizing.

Why this exists instead of codex resume: replaying a full rollout re-feeds every reasoning step, tool call, and tool output back into the context window. For long sessions that wastes the window on resolved turns and stale output. This skill selectively reconstructs only actionable context — the last compaction's surviving requests, recent user/assistant turns, the tool calls and files edited, and how the session ended — giving a fresh start with prior knowledge.

This is the Codex sibling of continue-claude-work. The two are deliberately split because the on-disk formats differ: Claude Code writes ~/.claude/projects/<encoded>/<session>.jsonl, Codex writes ~/.codex/sessions/.../rollout-*.jsonl with a different record schema. Use this skill for Codex sessions; use continue-claude-work for Claude Code sessions.

File Structure Reference

For the rollout directory layout, the record/payload schema, and the compaction format, see references/file_structure.md.

Workflow

Step 1: Extract Context (single script call)

Run the bundled extractor. It handles session discovery (via the shared _core), rollout parsing, noise filtering, and workspace state in one call: