Files
TheRock/rocm-systems/projects/amdsmi/.github/prompts/amdsmi-review-pr.prompt.md
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

123 lines
4.3 KiB
Markdown

---
description: "Review a GitHub pull request using the AMD-SMI Review Agent"
agent: "AMD-SMI Review Agent"
argument-hint: "<PR_NUMBER> [review-type ...] — e.g. 1234 style tests (or just the number for comprehensive). Add 'fast' to skip rebuttal. Add 'no-build' to skip build step. Add 'inherit' to use the orchestrator's model for subagents instead of the default (Sonnet 4.6)."
tools: [execute, read, edit, search, agent, web, todo]
---
Review a GitHub pull request using the AMD-SMI Review Agent.
## Arguments
- `$ARGUMENTS` contains: `<PR_NUMBER> [review-type ...]`
- First argument: PR number (required)
- Remaining: optional review types: `style`, `tests`, `docs`, `architecture`, `security`, `performance`, `build`, `skeptic`
- Special modifier: `fast` — skips the rebuttal round (comprehensive mode only)
- Special modifier: `no-build` — skips the build & install step
- Special modifier: `inherit` — makes all subagents inherit the orchestrator's model (the model you selected in the VS Code model picker) instead of their default (Sonnet 4.6)
- If no types specified, perform a **comprehensive** review (all subagents, with rebuttal)
## Process
### 1. Fetch PR Information
Use the PR number directly — `gh` resolves the repo from the local git remote:
```bash
gh pr view <PR_NUMBER> --json number,title,author,body,files,additions,deletions,state,baseRefName,headRefName,comments,reviews
```
Extract the `headRefName` (the PR branch name) — it is needed for the worktree step.
### 2. Set Up Worktree
Use the `amdsmi-using-git-worktrees` skill to check out the PR branch in an isolated worktree so the main checkout is not disturbed.
```bash
PR_NUM=<PR_NUMBER>
BRANCH=<headRefName from step 1>
MAIN_CHECKOUT=$(git rev-parse --show-toplevel)
WORKTREE_PARENT=$(dirname "$MAIN_CHECKOUT")
WORKTREE="${WORKTREE_PARENT}/rocm-systems-pr${PR_NUM}"
# Reuse existing worktree if already present
if [[ -d "$WORKTREE" ]]; then
cd "${WORKTREE}/projects/amdsmi"
git checkout "$BRANCH" 2>/dev/null && git pull origin "$BRANCH" --ff-only 2>/dev/null
else
cd "$MAIN_CHECKOUT"
git fetch origin "${BRANCH}:${BRANCH}" 2>/dev/null || git fetch origin "pull/${PR_NUM}/head:${BRANCH}"
git worktree add "${WORKTREE}" "${BRANCH}"
cd "${WORKTREE}/projects/amdsmi"
fi
```
If `git worktree add` fails (e.g. permissions), fall back to in-place work and note the fallback.
All subsequent steps (diff, build, file reads by subagents) operate from within this worktree.
### 3. Fetch the Diff
```bash
gh pr diff <PR_NUMBER>
```
### 4. Gather CI Evidence
Check for linked CI runs and fetch step-level data:
```bash
# Get CI check runs for the PR
gh pr checks <PR_NUMBER>
# For each failed or interesting run, get details
gh run view <RUN_ID> --json jobs
```
Find a recent baseline run on `develop` for comparison:
```bash
gh run list --branch develop --workflow <WORKFLOW> --limit 1 --json databaseId,conclusion
```
Compare step timings and status between PR run and baseline. Pass this evidence to the test and performance subagents.
### 5. Fetch Unresolved Comments
```bash
gh pr view $PR_URL --json comments,reviews,reviewRequests
```
Extract unresolved review comments for the unresolved comments analysis.
### 6. Dispatch Review
Invoke the **AMD-SMI Review Agent** with:
- PR metadata (title, author, files, +/- lines)
- The diff
- The review type(s) from `$ARGUMENTS` (or "comprehensive" if none)
- CI evidence (test results, step timings, baseline comparison)
- Unresolved comments
- The worktree path so subagents read files from the correct checkout
- If `inherit` was specified, tell the agent to have subagents inherit the orchestrator's model (ignore their frontmatter `model` field)
The agent will dispatch to the appropriate subagent(s) and produce a formatted review.
By default, comprehensive reviews include a rebuttal round (Round 2) with the skeptic subagent. If `fast` was specified, the rebuttal round is skipped.
### 7. Report Results
- Present the review inline
- Summarize the overall assessment
- List any blocking issues found
- If saving requested, use: `reviews/pr_{NUMBER}[_{TYPE}].md`
- The worktree at `rocm-systems-pr<PR_NUMBER>` persists for post-review investigation; cleanup follows the `amdsmi-restructure-commits` skill
## Examples
```
/amdsmi-review-pr 123
/amdsmi-review-pr 123 style
/amdsmi-review-pr 123 tests security performance
/amdsmi-review-pr 123 fast
```