Codex Rollout Report
Purpose
Render a Codex rollout transcript into a readable Markdown report. This skill is intentionally rollout-only: it reads rollout-*.jsonl transcript files and writes a report artifact. It does not use semantic audit logs, OTel logs, Grafana, Loki, hooks, Docker, network calls, or repository state.
Hidden reasoning is not available and must not be inferred. The report may include visible assistant messages, visible final answers, tool call metadata, tool results, and token-count records that are present in the rollout transcript.
Command
Use the bundled script:
python3 <skill-dir>/scripts/render_rollout_report.py --latest
Useful variants:
python3 <skill-dir>/scripts/render_rollout_report.py --rollout <path/to/rollout.jsonl>
python3 <skill-dir>/scripts/render_rollout_report.py --rollout <path/to/rollout.jsonl> --turn <turn_id>
python3 <skill-dir>/scripts/render_rollout_report.py --session <session_id>
python3 <skill-dir>/scripts/render_rollout_report.py --latest --out /tmp/codex/reports/
python3 <skill-dir>/scripts/render_rollout_report.py --latest --transcript-root <path/to/sessions>
Supported flags:
--rollout <path>: read a specific rollout JSONL transcript.--latest: select the newest rollout containing a reportable completed turn.--session <id>: filter by session id while searching rollout transcripts.--turn <id>: render a specific turn from the selected transcript.--transcript-root <path>: search a custom sessions directory instead of$CODEX_HOME/sessionsor~/.codex/sessions.--out <path-or-dir>: write to an exact Markdown file or to a directory.
Output
Default output directory:
/tmp/codex/reports/
On success the script prints:
path: <report-path>
open: file://<report-path>
The report contains:
MetadataPromptTimelineFinal AnswerFiles And Notes
Safety
- Keep reports under
/tmp/codex/reports/unless the user explicitly provides another non-repository output path. - Do not write reports inside detected Git worktrees.
- Do not modify repositories, hooks, Codex config, telemetry config, or background services.
- Treat token counts as rollout-derived and potentially incomplete.
- Treat visible assistant messages as visible transcript text, not hidden reasoning.
Validation
python3 -m py_compile <skill-dir>/scripts/render_rollout_report.py
python3 <skill-dir>/scripts/render_rollout_report.py --help