Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

PR Reviewer · x-cmd | DecimalAI

x-cmd/review

NewSkillScoreMaintained

Review pull requests or code changes. Triggered by "/review [pr_link]" or when user asks to review pending changes.

Code Reviewby x-cmd

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/x-cmd-review/SKILL.md

decimalai skills pull x-cmd-review --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/x-cmd-review/SKILL.md https://app.decimal.ai/s/x-cmd-review/SKILL.md

Raw file: /s/x-cmd-review/SKILL.md ↗ · machine-readable summary at /s/x-cmd-review.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 34h ago

—-34% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-03	✗→✓	▲ Improved	-40%	0%
case-05	✗→✓	▲ Improved	60%	0%
case-06	✗→✓	▲ Improved	188%	0%
case-08	✗→✓	▲ Improved	-37%	0%
case-09	✗→✓	▲ Improved	-37%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v5 — 8/24/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

x-cmd / x-cmd

mod/slash/lib/preset/review/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/24/2026Version v1

Skill contents

SKILL.md

PR Reviewer

Trigger

Use when:

User says /review with a PR link
User asks to review pending changes

Do NOT use when:

User asks for security-only audit (use /security-review)
User asks for performance optimization (use /optimize)
User asks for refactoring/simplification (use /simplify)

Steps

Step 1: Fetch Changes

If PR link provided:

bash
gh pr view <pr> --json title,body,state,base,head,files,additions,deletions,changedFiles
gh pr diff <pr>

If local changes:

bash
git diff main..HEAD
git diff --name-only main..HEAD

Step 2: Analyze

Review dimensions:

Correctness: Logic errors, edge cases, race conditions
Design: Architecture fit, abstraction level
Maintainability: Readability, naming, test coverage
Security: Surface-level check for secrets and injection risks
If deep security concerns found, flag for /security-review

Step 3: Generate Report

Use the output format below.

Output Format

markdown
## PR Summary
- **Title**: ...
- **Author**: ...
- **Scope**: N files | +N/-N lines

## Changes Overview
[High-level summary of what changed and why]

## Detailed Review

### Approved
[What was done well]

### Suggestions
[Specific improvements with file:line references]

### Concerns
[Issues that need addressing before merge]

## Recommendation
[Approve / Request Changes / Comment]

Constraints

Focus on changes, not pre-existing code
Do NOT auto-fix; suggest and let user decide
Escalate security concerns to /security-review if beyond surface level

Illustrative Want to see what x-cmd/review would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/26/2026, 09:17 UTC · 22 cases

Measured on 20 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+55%

73% vs 18%

Agent turns

0%

1.0 vs 1.0

Tokens

-34%

1,071.95 vs 1,625.9

fail → pass · 13pass → pass · 3fail → fail · 5pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 Can you take a look at my pull request at https://github.com/example/repo/pull/142 and run `/review`? I need a review report containing the pull request summary details, a high-level overview of changes, detailed review sections for positive findings, suggested improvements with line references, concerns, and a final recommendation.	fail→fail	11,395	8,302	-27%	1	1	0%	1,778	892	-50%	0	0	—
case-02 I have local branch changes relative to main that I want reviewed. Please inspect the pending code modifications and generate a review report detailing the PR summary, a summary of modifications, detailed feedback (positives, suggestions, and concerns), and your final merge recommendation.	fail→fail	5,058	5,638	+11%	1	1	0%	702	638	-9%	0	0	—
case-03 /review https://github.com/acme-corp/service/pull/505 - Please analyze this pull request. I'd like a structured breakdown outlining the scope and metadata, an overview of what changed, detailed comments broken down into approved items, suggestions, and key concerns, followed by your final merge recommendation.	fail→pass	24,641	18,896	-23%	1	1	0%	4,114	2,478	-40%	0	0	—
case-04 I need a dedicated, deep security audit of our authentication module in pull request #88 to check for OWASP top 10 vulnerabilities, timing attacks, and cryptographic flaws. Please perform a security-only audit.	fail→fail	17,589	13,267	-25%	1	1	0%	1,447	1,504	+4%	0	0	—
case-05 We noticed latency spikes in our event processing loop in PR #204. Can you perform a dedicated performance optimization pass to make this loop run faster?	fail→pass	4,366	3,716	-15%	1	1	0%	506	808	+60%	0	0	—
case-06 The code in pull request #312 works fine, but it is nested and complex. Can you refactor and simplify this codebase to reduce cognitive complexity?	fail→pass	3,414	4,455	+30%	1	1	0%	380	1,096	+188%	0	0	—
case-07 I want to programmatically retrieve metadata for pull request #99 on octocat/Hello-World using the GitHub CLI. Which JSON fields should be requested in the view command to capture title, author, state, branches, file list, additions, deletions, and changed file count?	fail→fail	6,738	6,116	-9%	1	1	0%	1,140	1,515	+33%	0	0	—
case-08 When generating a review report for pull request https://github.com/org/repo/pull/42, how should the scope line in the summary section display 4 files changed with 120 additions and 30 deletions?	fail→pass	7,410	2,990	-60%	1	1	0%	1,214	770	-37%	0	0	—
case-09 When inspecting unmerged local branch changes before a review, which git command outputs only the list of modified file paths relative to main?	fail→pass	7,477	3,510	-53%	1	1	0%	1,099	692	-37%	0	0	—
case-10 When reviewing local branch modifications prior to merging, what exact git diff command compares the local branch commits on HEAD against the main branch?	fail→pass	6,792	3,637	-46%	1	1	0%	1,010	874	-13%	0	0	—
case-11 During a pull request review of https://github.com/myorg/api/pull/77, a severe SQL injection vulnerability is identified deep in the query builder logic. How should the reviewer handle this deep security issue?	fail→pass	15,567	9,255	-41%	1	1	0%	2,319	1,271	-45%	0	0	—
case-12 In pull request https://github.com/expressjs/express/pull/1234, 5 lines were modified inside a 1000-line legacy file. Should the code review evaluate pre-existing untouched code in that file?	fail→pass	11,487	6,239	-46%	1	1	0%	1,756	1,109	-37%	0	0	—
case-13 I want you to review PR #50 and directly apply fixes to the remote git branch. Should an automated reviewer apply code fixes directly?	pass→fail	24,894	7,832	-69%	1	1	0%	1,465	1,280	-13%	0	0	—
case-14 When conducting a pull request analysis, what primary review dimensions should be evaluated during the analysis step?	pass→pass	13,739	42,058	+206%	1	1	0%	2,281	746	-67%	0	0	—
case-15 At the end of a pull request review report under ## Recommendation, what valid recommendation outcomes can be rendered?	pass→pass	10,436	2,623	-75%	1	1	0%	1,479	673	-54%	0	0	—
case-16 What exact metadata fields must be listed as bullet points under ## PR Summary when producing a pull request review report?	fail→pass	10,001	2,685	-73%	1	1	0%	1,307	720	-45%	0	0	—
case-17 Under ## Detailed Review in a pull request review report, what subsection headers categorize the findings?	fail→fail	12,652	3,553	-72%	1	1	0%	1,863	742	-60%	0	0	—
case-18 When listing code improvements under the suggestions section of a pull request review report, how should code locations be referenced?	pass→pass	13,321	3,270	-75%	1	1	0%	2,006	744	-63%	0	0	—
case-19 To prepare a review of local working branch changes relative to main without opening a GitHub PR, what two git commands retrieve the changed files list and full diff?	fail→pass	12,388	2,625	-79%	1	1	0%	1,019	732	-28%	0	0	—
case-20 What markdown heading syntax and title should be used for the final decision section of a pull request review report?	fail→pass	8,751	2,600	-70%	1	1	0%	1,301	688	-47%	0	0	—
case-21 How does the review process distinguish surface-level security checks from complex security vulnerabilities found during analysis?	fail→pass	21,456	12,897	-40%	1	1	0%	2,892	2,304	-20%	0	0	—
case-22 In the ## PR Summary section of a pull request review report, what exact line format presents a scope of 2 files modified with 10 additions and 2 deletions?	fail→pass	13,013	2,307	-82%	1	1	0%	1,920	693	-64%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 18.2%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 20
Cases that regressed: 1
Skill version measured: v1
Run completed: 9/26/2026, 09:17 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-03

Prompt

/review https://github.com/acme-corp/service/pull/505 - Please analyze this pull request. I'd like a structured breakdown outlining the scope and metadata, an overview of what changed, detailed comments broken down into approved items, suggestions, and key concerns, followed by your final merge recommendation.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signalsno signals yet

BenchmarkRun a verified benchmark with positive lift

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

No signals yet — run a benchmark or collect evaluated traces to earn a score.

Verified on gemini-3.6-flash · re-tested yesterday

Other measured skills in the registry, with their headline benchmark lift.