viettran-edgeai/codex_workflow/deployment-token-report

Deployment Token Report

<!-- codex-workflow-skill: deployment-token-report -->

Use this skill only as Archivist for an assigned deployment closure, after all assigned documentation updates, compact checks, and Git inspection are complete. Treat that closure state as sealed; the repository and closure evidence remain unchanged after reporting starts.

Use the supplied deployment ID unchanged. It must match [a-z0-9][a-z0-9_-]{0,63} and the main agent's deployment marker:

<!-- codex-workflow-deployment-start: <deployment_id> -->

Run the bundled scripts/report_tokens.py with --deployment-id and --format markdown. Let it use CODEX_THREAD_ID to identify this Archivist rollout, resolve its parent main-agent thread, find the exact marker in assistant message text there, and read only metadata and token-count fields beneath ~/.codex/sessions/. Accept the marker when surrounded by Markdown or explanatory prose. Exclude guardian sessions.

Return only the script's six-column Markdown table verbatim to the main agent, with no pricing, estimates, inferred usage, or additional statistics. Treat a script failure or incomplete-evidence warning as the report result and return the limitation verbatim.

The required table template is exactly:

| Agent | Quantity | Rollouts | Cached input | Input | Output |
| --- | ---: | ---: | ---: | ---: | ---: |
| <agent role> | <count> | <count> | <tokens> | <tokens> | <tokens> |

Keep the columns exactly as shown and preserve every data row emitted by the script, including its final main agent row.

Interpret Input as total input tokens, including the cached-input subset, and Rollouts as model generations with a last_token_usage record. Stop totals when the script starts, excluding Archivist's post-tool final response and the main agent's later final response.