codex-session-exporter

# Codex Session Exporter Use the bundled script for path resolution and HTML export. Do not rewrite the parser ad hoc. ## Workflow 1. Obtain the user's input: an existing JSONL path or a complete UUID session ID. 2. Determine the output path. Use a user-specified path exactly; otherwise, use `codex-session-<session-id>.html` in the current working directory. 3. Run: ```bash python3 "$SKILL_DIR/scripts/export_session.py" "$SESSION_OR_ID" \ --output "$OUTPUT_HTML" ``` 4. Return both absolute paths printed by the script: `Session source` and `HTML report exported`. `SKILL_DIR` is the current skill's real directory. Do not assume the caller runs from the repository root. ## Session ID lookup Accept only a complete UUID, not a prefix or fuzzy ID. Search these locations in order by default: - `$CODEX_HOME/sessions/**/rollout-*-<session-id>.jsonl` - `$CODEX_HOME/archived_sessions/**/rollout-*-<session-id>.jsonl` - Use `~/.codex` when `CODEX_HOME` is unset If the session is under an isolated benchmark `agent-home/.codex`, pass that home explicitly: ```bash python3 "$SKILL_DIR/scripts/export_session.py" "$SESSION_ID" \ --codex-home "/absolute/path/to/agent-home/.codex" \ --output "$OUTPUT_HTML" ``` If no session is found, report the searched CODEX_HOME. If multiple files match, do not choose one; ask the user for a complete path. ## Constraints - Use Python 3.11+ and do not install dependencies. - Do not modify the original JSONL. - Do not attempt to decrypt reasoning; show only plaintext summaries already present in the rollout. - Preserve the report's ability to download the original JSONL. The report contains the complete session text; do not upload or share it without user authorization. - For export-only requests, do not invoke the benchmark parser or analysis scripts. Use `codex-session-analyzer` when the user needs an explanation of failures, timing, or tool paths.