Review Pr

Review AFD Plugin Pull Requests

Review like a maintainer: direct, selective, and focused on issues that CI does not prove. Prefer a few high-confidence findings over exhaustive commentary. Zero findings is a valid result.

Quality contract

Make every finding:

• Correct: prove a reachable failure, not a suspicion.
• Prioritized: lead with merge blockers and high-impact defects.
• Actionable: identify the smallest safe fix direction.
• Evidence-based: cite code, tests, docs, CI, or measurements.
• Concise: avoid review templates and repeated summaries.
• Calibrated: match severity to user and maintainer impact.

Do not report unrelated backlog, style already enforced by pre-commit (ruff, mypy, SPDX headers, markdownlint, DCO), or a missing test that would not protect changed behavior.

Select the input and depth

Use vllm-project/afd-plugin as the base repository. Accept its forks and local checkouts; use another skill for unrelated repositories.

Default to maintainer brevity. A detailed or audit request expands coverage and lists path:line findings, but keeps the same confidence and severity bar.

Reference guide

Load references in review order: process, one primary module contract, matching feature overlays, evidence checks, then delivery. Every file is linked directly below; do not load unrelated module references.

Each module contract links to a design page under docs/design/module/. For branch-specific behavior, inspect the matching design page in the reviewed checkout first; the frozen head's design pages are authoritative for that review. If docs and live code disagree, verify the code/tests and report the drift.

Review process

Primary module contract

Feature-design overlays

Evidence and quality checks

Delivery and reviewer coordination

Workflow

1. Freeze and report the snapshot

Pin the base and head before reading source or running validation. Within 60 seconds, report the pinned head, CI, mergeability, and preliminary findings in the host conversation. Do not wait for CI or post this update to GitHub.

If the target changes while fetching, discard the evidence and retry once. If it changes again, report the churn and wait for a stable target.

For a trusted PR head, materialize the pinned head in an isolated detached worktree. A worktree freezes identity but is not a security sandbox. Treat fork heads as untrusted unless the user and environment policy explicitly establish otherwise: execute them only in a disposable, secret-free sandbox with restricted filesystem, network, and resources; without one, use static SHA-addressed reads and CI evidence only. For a local review, freeze the committed, index, worktree, and NUL-safe in-scope untracked contents. Follow review-execution.md for trust gates, state fingerprints, and byte-for-byte staleness checks.

2. Build the diff census

Group files into production code, tests, docs, configuration, build/CI, and vendored artifacts (for example afd_plugin/connectors/npu/bin/). Map each changed production file and test group to the PR goal. Compare the title/body claims with the actual diff; use linked issues only when they define the contract or reproduction.

Mark unrelated scope and unexplained generated artifacts. Do not infer behavior from the PR description without tracing the live code.

3. Route from the live behavior

Trace each claimed behavior through the changed producer to its live consumer, then use design-contracts.md and review-routing.md to select one primary module contract, a second only for a real documented cross-boundary call path, and every matching feature and evidence overlay. Treat titles and paths as hints; live behavior and the frozen head's current design metadata are authoritative. For docs-, tests-, or CI-only changes, route to the production contract they protect or use only the applicable evidence checks.

4. Run the blocker scan

Apply every category in general-checks.md before lower-priority comments.

For each changed value or behavior, trace:

vLLM engine/config ingress (--additional-config {"afd": ...})
  -> parse_afd_config / validate_afd_config -> platform/worker selection
  -> role worker/runner startup -> connector handshake
  -> attention<->FFN transfer -> model role forward (remote proxy)
  -> sampling/output in the owning role -> shutdown/teardown

Cover every applicable GPU/NPU, graph/eager, DBO on/off, 2A2F/2A1F, runner V1/V2, colocation/disaggregation, and AFD-on/AFD-off path. Patched code must stay transparent when AFD is inactive. Search bounded callers and sibling implementations rather than assuming the changed hunk is the only path.

5. Apply module, feature, and patch contracts

Apply the reference set selected in step 3 and any matching repo-local skill. Read the exact module and feature pages in the frozen head, including status, ownership, primary_code_paths, dependencies, and last_reviewed. Candidate or draft rules are questions, not blockers, unless current code, tests, or AGENTS.md enforce them. Inspect both sides of every config, registration, connector, rank-mapping, and patch boundary.

For every change under afd_plugin/compat/, enforce the AGENTS.md patch contract: pinned upstream refs, full upstream copy with marked AFD differences, identical signatures, Patch reason/Patch functionality comments, and CPU-safe test coverage.

When a diff adds or expands a helper, class, fallback, compatibility branch, or public behavior, run a subtraction pass: remove out-of-scope behavior and check whether each new abstraction can be deleted, merged, moved, or inlined.

6. Verify the changed path

Before each validation group, verify the frozen SHA plus the tracked, index, untracked, and ignored-file fingerprint, or recreate a pristine snapshot. On a trusted head or inside the required sandbox, run an import/version preflight, then the narrowest relevant tests and low-cost static checks. Bind every result to the head SHA, snapshot fingerprint, and environment fingerprint. Never run imports, tests, builds, hooks, or repo-configurable tooling from an untrusted head on the reviewer host.

• Treat CI as status evidence; inspect only the first overlapping failure. Buildkite test-ready runs unit tests plus the DeepSeek-V2-Lite E2E gate; NPU has no in-repo CI, so NPU evidence is author-provided or manually run.
• For docs-only changes, use diff hygiene, links checks, and bounded live contract verification instead of dependency setup or pytest.
• For hardware-dependent paths, run available static/CPU checks (pytest -m "not gpu and not vllm_runtime") and name the exact GPU/NPU gap. Never simulate device evidence.
• For performance or accuracy claims, require comparable base/head runs with the same environment, workload, warmup, repetitions, and graph/DBO mode.

Stop when each changed semantic path has a supported finding or an explicit no-issue conclusion. Do not search further only to increase confidence.

7. Consolidate and deliver

Verify each finding against the current diff, deduplicate by root cause, and order by severity.

Re-read the remote head and reverify or recreate the pristine validation snapshot immediately before delivery. If either changed, mark the review stale and restart from the new snapshot.

Return findings first. Use delivery-style.md to keep them direct and brief. Each finding must include an exact path:line, trigger or call path, current behavior, impact, and smallest fix direction. If there are no findings, say so briefly and name material validation gaps.

Keep the review read-only unless the user explicitly authorizes posting. Do not submit APPROVE, COMMENT, or REQUEST_CHANGES, add labels, edit code, or push commits as an implied part of review.

8. Optionally request focused owner reviews

Only when the user asks to identify or request reviewers, read review-requests.md. Rank path-matched CODEOWNERS against the frozen module contract's owners and documented governance expertise; propose focused reviewers with an explicit contract rationale.

Identifying or suggesting reviewers is read-only. Requesting reviewers or posting @mention comments changes external state and requires explicit user authorization. When authorized, recheck the head, deduplicate existing requests, and post at most one consolidated comment. Do not infer this permission from a request to review the code.

Related skills

• run-e2e — execute the DeepSeek-V2-Lite / Qwen3 MoE / Qwen3.6 E2E gates; use when the review needs active GPU or NPU evidence.
• adapt-model — author-side workflow for adapting a new model family; suggest to contributors alongside the model-addition checklist.
• upgrade-npu / upgrade-gpu-version — the pinned-ref upgrade workflow; a PR that moves vLLM or vLLM-Ascend pins should follow them.