Surface runtime model + token usage in published review summaries

[Copilot]

Review outputs currently dropped runtime metadata collected by ACP, so maintainers could not see which model produced a review or the token cost of the run. This change propagates structured runtime metadata end-to-end and renders it in the final summary body (including consolidated GitHub review bodies) with explicit unavailable states.

• Runtime metadata propagation

  • Added runtime fields to agent/review result flow so model + usage survive parsing and publishing boundaries.
  • AgentRunResult now carries model; ACP runtime captures response-reported model when present.
  • SkillResult now carries attached runtime metadata (model, input_tokens, output_tokens, total_tokens).

• Usage aggregation at runner boundary

  • CodingAgentReviewRunner aggregates runtime data across all agent calls in a review run, including retry paths.
  • Token totals are computed per metric only when that metric is available for all calls; otherwise rendered as unavailable (avoids misleading partial totals).
  • Multiple runtime models are labeled explicitly when retries/reporting differ.

• Summary rendering updates

  • format_review_note(...) now renders a compact runtime footer line near bot attribution.
  • Applied in both platform and debug publishers, so standalone summary comments and consolidated review bodies share the same runtime line.
  • Existing bot attribution, skill fingerprint, and hidden metadata comment remain intact.

# rendered footer example
"Model: provider/model · Tokens: input 12,345 / output 678 / total 13,023"

# when data is incomplete/missing
"Model: unavailable · Tokens: input unavailable / output unavailable / total unavailable"

[github-actions]

Code Review

Clean runtime metadata propagation end-to-end with solid test coverage. Low overall risk; one minor type-check defensiveness improvement suggested.

Severity: Critical 0 / High 0 / Medium 0 / Low 1
Inline comments posted: 1

Generated by whhe/code-review-bot · Skill fingerprint: 43b5df0c

[whhe]

@copilot fix the comment #6 (comment)

[Copilot]

@copilot fix the comment #6 (comment)

Fixed in commit 7f83bde. CodingAgentReviewRunner now excludes booleans when aggregating token counts (isinstance(value, int) and not isinstance(value, bool)), and I added a regression test covering bool usage values.

[github-actions]

Code Review

Clean implementation that propagates ACP runtime model and token usage end-to-end with well-designed aggregation semantics and good test coverage. No new issues found; the previously flagged bool/int edge case is fixed.

Severity: Critical 0 / High 0 / Medium 0 / Low 0
Inline comments posted: 0

Generated by whhe/code-review-bot · Skill fingerprint: 43b5df0c