| name | cross-review |
| description | Run an OPTIONAL, advisory cross-vendor code review of the current branch's diff vs main using a configurable second-opinion backend (Azure Foundry next-gen, GitHub Models, or Copilot CLI - all OAuth, all keyless). Explicit-only; run only when invoked via /cross-review or when the user explicitly asks for a cross-vendor review, never autonomously. |
| argument-hint | ["profile-name"] |
| disable-model-invocation | true |
| allowed-tools | Bash |
Cross-vendor review
You are running a cross-vendor code review on the current branch's diff vs main, using the reviewer profile the user named (or the default from profiles.json). This is a second opinion from a model in a different vendor lineage. It is advisory only: surface findings, do not apply changes.
Run the review
!bash ${CLAUDE_SKILL_DIR}/scripts/review.sh "$1" 2>&1
Present the result
The script printed the findings above and saved a markdown copy under flow/reviews/. In your reply:
- Surface the findings clearly to the user, with the saved file path so they can re-read it.
- Do NOT modify any code based on what the reviewer said. The reviewer is a second opinion, not an instruction.
- If the user asks you to act on a specific finding, treat it as a separate task and apply your own judgement; the reviewer may be wrong.
If the script errored
Common causes (the printed error points at one of these):
profiles.json missing: copy .claude/skills/cross-review/profiles.example.json to profiles.json and edit endpoints/models.
- Authentication:
az login (Foundry), gh auth login (GitHub Models), copilot auth login (Copilot CLI).
- Tool not installed:
az, gh, copilot, jq, or curl missing.
- No diff vs
main: the branch has no changes to review.
Tell the user the specific error from the script output and the fix; full setup is in docs/cross-vendor-review.md.