GaoSSR/codex-export-skill: Codex Skill for exporting sessions as LLM-ready Markdown transcripts · GitHub

Export Codex sessions as clean, LLM-ready Markdown.

🌐 English | 🇨🇳 简体中文

export is a Codex Skill that brings Claude Code's /export workflow to Codex users. It turns local Codex session history into clean Markdown transcripts, making conversations easier for LLMs to parse, review, and analyze.

Requirements:

• Codex with Skill support
• Python 3.10 or newer available as python3
• Node.js/npm for npx skills

npx skills add GaoSSR/codex-export-skill --agent codex -g -y --copy

Restart Codex after installation so the $export trigger is discovered.

If you installed an earlier version, remove stale copies first. Otherwise an old ~/.codex/skills/export can shadow the newer install created by npx skills.

rm -rf "${CODEX_HOME:-$HOME/.codex}/skills/export" "$HOME/.agents/skills/export"
npx skills add GaoSSR/codex-export-skill --agent codex -g -y --copy

Restart Codex after reinstalling.

Inside Codex, ask:

$export export the current session to Markdown

The Skill writes the transcript to codex-session-exports/ under the active workspace and replies with the absolute file path plus a short export summary.

More examples:

$export list recent Codex sessions
$export export session <session-id> to Markdown
$export export this session with tool logs

No extra shell commands are required after installation.

Session export is useful when you want another model pass to review how a conversation went, identify repeated mistakes, and turn those lessons into durable project rules such as AGENTS.md.

Markdown is the default output because LLM conversations already use Markdown heavily. The transcript stays easy to read, diff, archive, and feed back into another model for analysis.

• One-command installation: install as a Codex Skill with npx skills add.
• Conversation-native usage: use $export directly inside Codex after installation.
• Markdown-first output: produces clean transcripts designed for humans and LLMs.
• Current-session selection: prefers the active Codex conversation when available.
• Workspace-aware fallback: falls back to the latest session for the current workspace, then latest globally.
• Privacy-conscious defaults: excludes system prompts, developer instructions, AGENTS context injection, environment context injection, reasoning records, and tool logs by default.
• Path redaction by default: Markdown metadata shows only local source basenames unless full source paths are explicitly requested.
• Optional tool-log export: include tool logs only when you explicitly ask for them.

By default, the export includes:

• visible user messages
• visible assistant messages
• session metadata such as session id, source file, cwd, timestamps, originator, and CLI version

Local source file paths and cwd metadata are redacted by default. The exported Markdown keeps only basenames such as rollout-...jsonl and the workspace folder name.

By default, the export excludes:

• system prompts
• developer instructions
• AGENTS or project-doc context injection
• Codex Skill context injection such as <skill>...</skill>
• environment context injection
• encrypted or summarized reasoning records
• tool calls and command output

Tool calls and command output are exported only when you explicitly ask the Skill to include tool logs. If Codex Skill context appears inside a visible message, the exporter redacts that internal <skill>...</skill> block while preserving the surrounding user text.

Advanced script users can pass --json for machine-readable output, or --show-paths when they intentionally need full local source paths in Markdown metadata.

When you do not specify a session, the Skill tries to export the current Codex conversation first. If the current conversation id is unavailable, it falls back to the latest session recorded for the active workspace, then to the latest session globally.

To export a specific session, ask by session id:

$export export session <session-id> to Markdown

The long-term target is native Codex CLI support for /export, with Markdown as the default transcript format. This repository keeps the workflow usable as a Skill until native support is available upstream.

Issues and pull requests are welcome. Please keep changes aligned with the core contract: a simple Codex Skill surface, Markdown-first output, and conservative privacy defaults.

Licensed under the Apache License 2.0.

Copyright 2026 GaoSSR.

This is not an official OpenAI project.