Getting it into your agent
There is no command for this one: it runs only inside a plugin, and the catalogue could not identify which plugin ships it. The source is linked below.
Wrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-rocm-pr-review)<a href="https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-rocm-pr-review"><img src="https://agentmods.dev/badge/skills/shen-shanshan/vllm-dev-skills/vllm-rocm-pr-review/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-rocm-pr-review"><img src="https://agentmods.dev/badge/skills/shen-shanshan/vllm-dev-skills/vllm-rocm-pr-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00137 | $0.03225 |
| Opus 5 | $0.00068 | $0.01613 |
| Sonnet 5 | $0.00027 | $0.00645 |
| Haiku 4.5 | $0.00014 | $0.00323 |
Grade A, and why
vllm-rocm-pr-review scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 13d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 183 lines — stays where its author put it; the contents beside it link to each section on GitHub.
vLLM ROCm PR Review
Review a vllm-project/vllm PR through an AMD/ROCm lens, then write a concise Chinese review report to ./outputs/pr-<NUMBER>-review.md.
Knowledge sources (kept in references/):
| File | Content | When to read |
|---|---|---|
references/vllm-rocm-review-rules.md |
Curated rule catalog adapted from the ATOM/aiter review skills to vLLM's ROCm surface | The mandatory checklist (Step 6) |
references/upstream/atom-review-pr-skill.md |
Full upstream ATOM review-pr skill | Deep reviews / new rule discovery |
references/upstream/aiter-review-pr-skill.md |
Full upstream aiter review-pr skill | Deep reviews / new rule discovery |
references/upstream/mori-readme.md + mori-structure.json |
mori (AMD comm library) README + repo tree snapshot | PRs touching mori / MORI-EP / moriio |
references/example-report.md |
A completed Chinese review report as style anchor | Before writing the report |
Context economy: for quick reviews, read only the rule categories mapped by the PR type (Step 4) plus the core discipline section of the rules file. For deep reviews (Tier-1 backbone files touched, mori changes, perf claims), read the full rules file.
Workflow
- Fetch PR data —
scripts/fetch_pr_data.py - ROCm relevance gate
- Semantic understanding (5 questions)
- PR-type classification → mandatory rule categories
- Backbone-file risk assessment
- Apply the rule checklist
- AI-code diagnostic + blind-spot check
- Write the report to
./outputs/and tell the user the path
Step 1: Fetch PR Data
python3 /Users/shanshan-shen/.claude/skills/vllm-rocm-pr-review/scripts/fetch_pr_data.py <PR_NUMBER> \
--output /tmp/vllm_rocm_pr_<PR_NUMBER>.json --include-ci
The JSON contains:
| Key | Content |
|---|---|
pr |
PR metadata: title, body, author, state, labels, additions/deletions, branch names, linked reviews |
diff |
Full unified diff of all changed files |
files |
Per-file stats: filename, status, additions, deletions, patch |
issue_comments |
General discussion comments |
review_comments |
Inline code review comments |
reviews |
Top-level reviews (approvals, change requests) |
checks |
CI check status (only when --include-ci was passed; absent = CI status unknown) |
What ships with it
30 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- outputs/pr-29304-review.md 14 KB
- outputs/pr-38434-review.md 7.3 KB
- outputs/pr-51004-review.md 11 KB
- outputs/pr-51040-review.md 11 KB
- outputs/pr-51171-review.md 10 KB
- outputs/pr-51563-review.md 8.9 KB
- outputs/pr-51705-review.md 12 KB
- outputs/pr-52033-review.md 9.8 KB
- outputs/pr-52882-review.md 11 KB
- outputs/pr-52921-review.md 8.6 KB
- outputs/pr-53161-review.md 11 KB
- outputs/pr-53407-review.md 9.4 KB
- outputs/pr-53598-review.md 11 KB
- outputs/pr-53664-review.md 9.5 KB
- outputs/pr-53698-review.md 9.6 KB
- outputs/pr-53821-review.md 7.9 KB
- outputs/pr-53838-review.md 9.5 KB
- outputs/pr-53987-review.md 10.0 KB
- outputs/pr-54682-review.md 11 KB
- outputs/pr-54787-review.md 9.3 KB
- outputs/pr-54855-review.md 9.5 KB
- references/example-report.md 14 KB
- references/upstream/.update-metadata.json 1.6 KB
- references/upstream/aiter-review-pr-skill.md 58 KB
- references/upstream/atom-review-pr-skill.md 49 KB
- references/upstream/mori-readme.md 15 KB
- references/upstream/mori-structure.json 61 KB
- references/vllm-rocm-review-rules.md 25 KB
- scripts/fetch_pr_data.py 8.0 KB runs code
- scripts/update_skill.py 12 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 13d ago First seen · 183 lines · 137 tokens per session scan A 0c49e46d8ea2
vllm-rocm-pr-review is a skill published in the GitHub repository shen-shanshan/vllm-dev-skills (17 stars, last pushed 3d ago), licensed Apache-2.0. It adds 137 tokens to every session and 3,225 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
open-code-review
Performs AI-powered code review on Git changes using the ocr CLI from alibaba/open-code-review. Use when the user asks to review code, review a pull request, review staged/unstaged changes, review a commit, or compare branches for code quality issues. Produces line-level review comments and can automatically apply…
continual-learning
Nightly refinement of an existing per-repo review-style prompt using this reviewer's own finding outcomes. Read confirmed (resolved-by-commit / thumbs-up) and dismissed (thumbs-down) findings, promote the bug patterns the team actually fixes, demote the false-positive patterns, reconcile against the current prompt…
camsnap
Capture frames or clips from RTSP/ONVIF cameras. Grabs snapshots, video clips, and motion events from IP cameras, security cameras, and video streams. Use when the user wants to take a snapshot from a camera, record a clip from an RTSP stream, monitor motion on a security camera, discover ONVIF devices on the network…
learn-from-sage
Detection-gap (miss) analysis for Code Review Sage. Learn from shipped fixes, acted-on human comments, and design outcomes to close reviewer blind spots. Inline during review stages a candidate; a human triggers a one-shot AI consolidation into the live ruleset.
meta-codereview-current-diff
Read the current uncommitted diff, run three independent reviewers (safety + tests-coverage + style) in parallel, then arbitrate a single BLOCK / BLOCKWITHOVERRIDE / PASSWITHNOTES verdict. Use before commit when you want a multi-perspective second-opinion instead of a single-reviewer agent loop.
omh-image-cards
This is a Hermes-native img-summary workflow skill.