Borrowing it
Nothing to install: this file belongs to ROCm/aiter. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/ROCm/aiter/main/.claude/skills/review-pr/SKILL.mdgit clone --depth 1 https://github.com/ROCm/aiterWrote 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/rocm/aiter/review-pr)<a href="https://agentmods.dev/skills/rocm/aiter/review-pr"><img src="https://agentmods.dev/badge/skills/rocm/aiter/review-pr/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/rocm/aiter/review-pr"><img src="https://agentmods.dev/badge/skills/rocm/aiter/review-pr.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 4 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Anti-Refusal · line 932 Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
- high System Prompt Leakage · line 1287 Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.Fix: Remove any instructions that reveal, print, or output system prompts or internal rules. System instructions should never be exposed to end users.
- medium Excessive Agency · line 1075 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Excessive Agency · line 989 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00171 | $0.09996 |
| Opus 5.5 | $0.00068 | $0.03998 |
| Sonnet 5.5 | $0.00034 | $0.01999 |
| Haiku 4.5 | $0.00017 | $0.01000 |
Grade A, and why
review-pr 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 1mo 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.
Copies of this mod
1 near-identical copy found in the catalogue:
- review-pr — 100% identical, 696 lines differ
How it starts
The opening of the file, as written. The whole thing — 501 lines — stays where its author put it; the contents beside it link to each section on GitHub.
aiter PR Review — advisory tier
This skill supplies hints to a human reviewer. Its judgement is stochastic and never blocks a
merge. Only a reproducible blocker from an explicitly supplied, head-matched
validation_report.json may be used as a deterministic gate.
Promotion bar
The two conditions under which this could stop being advisory, and why neither
holds yet: rules.md § Promotion bar.
Step 1 — Fetch
# Everything Step 1 does is in fetch.sh; read its output, then keep the printed $WORK.
"$(git rev-parse --show-toplevel)/.claude/skills/review-pr/fetch.sh" "$@"
Read the diff and PR body before proceeding.
Step 1b — Derive the applicable rules, and collect the evidence they need
Step 1b writes its artifacts into $WORK. Read them; each explains its own output,
so what follows is the map, not the manual.
| file | what it answers | the trap it exists for |
|---|---|---|
rules_expanded.txt |
the full text of exactly the rules this diff derives | reading all 51 means attending to none |
applies.txt |
whether the diff still applies to the merge target | a stale PR's CI result describes a tree that moved |
merge_target.txt |
where the base tree is checked out — read base files from there | grepping the local worktree answers about the wrong branch |
guards.txt |
each deleted assert/check: moved, returned changed, or gone | "it came back" and "it was weakened" look identical |
siblings.txt |
a variant of a changed function still carrying a changed line | A1's sibling is in the same file, not another file |
flydsl_bounds.txt |
FlyDSL buffer/descriptor bounds not tied to the tensor's real extent | B2 is tl.load without a mask and FlyDSL has no tl, so this class had no rule at all |
aot_pairing.txt |
new ops-side contracts aiter/aot/flydsl/ was not taught |
#4397 wired stage2 into AOT and missed stage1; it drifted for 32 commits |
symbols.txt |
first-party imports that do not resolve against the merge target | a rebase signal, not invented code — #4994's import was valid when written |
twins.txt |
which existing file each new file was copied from | the defect is the asymmetry between them, not the copy |
test_quality.txt |
assertion count, tolerances, shapes of added tests | zero assertions may mean a helper asserts — read before firing |
kernel_tests.txt |
new kernels for which no test pytest collects was added | a benchmark is the shape these ship instead of a test |
ci_coverage.txt |
whether a CI job will ever run the added tests | HK6 is satisfied by a file in a directory nothing scans |
perf_claims.txt |
every number claimed, and which name no baseline | a signed delta and a before | after table already carry theirs |
struct_abi.txt |
structs whose pinned layout this diff shifts | the assertions exist to force a code-object rebuild |
comment_only.txt |
the non-prose lines of a comment-dominated diff | 7963 lines that reduce to none (aiter#4062) |
evidence.txt |
how a removed guard is handled on head — only written when a guard or signature changed | prose telling you to grep was read and not acted on (#5143) |
What ships with it
26 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.
- _apply_refutation.py 2.2 KB runs code
- _collect.py 3.8 KB runs code
- _gates.py 986 B runs code
- _lib.py 5.5 KB runs code
- _notify.py 3.8 KB runs code
- _publish.py 6.0 KB runs code
- authorized.txt 897 B
- check_prompts.py 9.4 KB runs code
- fetch.sh 48 KB runs code
- lost_review_test.py 7.3 KB runs code
- lost_review.py 3.2 KB runs code
- MAPPING.md 2.2 KB
- preflight.sh 5.8 KB runs code
- prompts/README.md 1.6 KB
- prompts/refuter.md 3.3 KB
- prompts/worker.md 12 KB
- queue_gate_test.py 15 KB runs code
- queue_gate.py 9.0 KB runs code
- render.sh 2.0 KB runs code
- rules.md 66 KB
- run_one.sh 14 KB runs code
- RUNNER-SETUP.md 5.4 KB
- selftest.sh 12 KB runs code
- triage.py 139 KB runs code
- watchdog_test.py 16 KB runs code
- watchdog.py 7.4 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.
- 1mo ago Changed · -872 lines 162dc6065c4c
- 1mo ago Changed · +740 lines · +110 tokens per session c4800ef8df9f
- 2mo ago First seen · 633 lines · 61 tokens per session scan A 0d1d39c6ccbe
review-pr is a skill published in the GitHub repository ROCm/aiter (598 stars, last pushed today), licensed MIT. It adds 171 tokens to every session and 9,996 once invoked, about $0.0007 per session on Opus 5.5. 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
autoreview
Structured code review through the shared OpenClaw agent-skills installation.
omh-code-review
This is an OMH code-review workflow skill, projected for Agent Skills hosts (Claude Code, Codex, Cursor, opencode, OpenClaw, pi).
revdiff-plan
Review the last Codex assistant message (plan, analysis, or proposal) with inline annotations in a TUI overlay. Extracts the most recent response from Codex rollout files and opens it in revdiff for review and annotation. Activates on "revdiff-plan", "review plan with revdiff", "annotate plan", "review last response"…
explore-unknowns
Guide the user through a quadrant walk that maps the unknowns of a task — open by listing the known knowns, then work through known unknowns, unknown knowns, and unknown unknowns one stage at a time, ending with a complete four-quadrant map in the user's hands. Use when a request is ambiguous or underspecified, the…
re0-merge
Review and land an external contribution the way this suite does: gate it against the thesis, land it with the author's credit intact, complete a new skill rather than merging it raw, then approve, credit, and explain before closing. Use when reviewing a pull request, as any collaborator or maintainer, not only the…
asplos-writing-style
Use when drafting or revising ASPLOS prose — engineering the first two pages that rapid review actually reads, stating the cross-layer insight as one quotable sentence, structuring per-layer mechanism sections, writing bounded claims, and fitting everything into 11 self-contained pages that include figures and…