Description
Include the model and token usage in the final review summary so maintainers can understand which model produced the review and how many tokens the run consumed.
Current behavior
The ACP runner already collects token usage from the agent response and writes it to the application log. However, this information is discarded before the review result reaches the publisher.
As a result, the final GitHub comment or review body shows the findings, severity counts, bot attribution, and skill fingerprint, but not the model or token consumption.
Expected behavior
The final published review summary should include a compact runtime line, for example:
Model: provider/model · Tokens: input 12,345 / output 678 / total 13,023
Acceptance criteria:
- Display the model that actually handled the review when the runtime reports it.
- Display input, output, and total token counts.
- Aggregate usage across all agent calls belonging to the review, including JSON repair retries.
- Clearly mark unavailable model or usage information instead of presenting misleading zero values.
- Include the information in both standalone summary comments and consolidated GitHub approval or request-changes review bodies.
- Do not expose API keys, provider endpoints, or other sensitive runtime configuration.
- Preserve the existing bot attribution, skill fingerprint, and hidden review metadata.
Suggested direction
Propagate structured model and usage metadata alongside the parsed review result, aggregate it at the review-runner boundary, and render it as a compact line near the existing attribution footer.
Description
Include the model and token usage in the final review summary so maintainers can understand which model produced the review and how many tokens the run consumed.
Current behavior
The ACP runner already collects token usage from the agent response and writes it to the application log. However, this information is discarded before the review result reaches the publisher.
As a result, the final GitHub comment or review body shows the findings, severity counts, bot attribution, and skill fingerprint, but not the model or token consumption.
Expected behavior
The final published review summary should include a compact runtime line, for example:
Model: provider/model · Tokens: input 12,345 / output 678 / total 13,023Acceptance criteria:
Suggested direction
Propagate structured model and usage metadata alongside the parsed review result, aggregate it at the review-runner boundary, and render it as a compact line near the existing attribution footer.