- Home
- Skills
- Code Quality & Review
- Code Review and Quality: Multi-Axis
More screenshots
Works with the AI Agents you already use
Code Review and Quality: Multi-Axis
Review a change in three passes — intent, behavior under stress, structural shape — with severity-labeled comments and an explicit merge verdict, so the codebase is healthier after every merge.
$5
Code Review and Quality: Multi-Axis
Example session with this skill installed
Review this pull request. Title: "Add CSV export to the reports page." The change adds an endpoint that runs the report query with no row limit, loads all rows into memory, and builds the CSV by joining fields with commas. Report titles are user-supplied. There is no check that the requesting user owns the report. One test was added; it exports a two-row report and checks the first line.
- Read your context and instructions
- Compiled the code review
Intent, in one sentence: let users download a report as CSV. Stated — proceeding to the diff.
Blocker: Report titles are user-supplied and are written into cells unescaped. A title starting with = or + executes as a formula when the file is opened in a spreadsheet — CSV injection, pointed at whoever downloads the export. Escape leading =, +, -, @ (and quote fields containing commas or newlines, which currently also corrupts the file's structure).
Blocker: The endpoint never checks that the requester owns the report. Any logged-in user can export any report by ID. Add the ownership check used by the report view path.
Should fix: The query has no row limit and loads everything into memory. A large report will take the worker down with it. Stream the response or page the query with a stated cap, the way the PDF export already does.
Should fix: The test exports two rows and checks the first line only. The cases that matter are exactly the untested ones: a title containing a comma, a title starting with =, and a user requesting someone else's report. Add those three.
Nit: csv_string is built in the controller; the existing export helper is the natural home for the escaping logic too.
Verdict: Request changes. Flip conditions: the two Blockers fixed, the three test cases added. The memory issue can ship behind the stated cap if the follow-up is filed with an owner.
Connects securely to your tools. The creator never sees your data.
About this skill
Code review usually fails one of two ways: the rubber stamp that lets a retry loop with no cap into production, or the style court that buries the one real finding under thirty nits until the author stops reading. This skill runs review as a three-pass protocol with a severity contract. Pass 1 reads the description, task, and tests before the diff — if you cannot state the change's intent in one sentence, it goes back before any code is read. Pass 2 stresses behavior: the paths tests miss, trust boundaries treated as hostile, and the load patterns that only hurt at scale, each finding quantified instead of felt. Pass 3 weighs the structural shape left behind — patterns followed or invented, layers respected, complexity earning its keep, dependencies justified. Every comment carries exactly one label (Blocker, Should fix, Nit, Note) from a finding-to-severity table, comments are ordered by leverage, and the review closes with an explicit verdict: Approve, or Request changes naming exactly what would flip it. Includes a comment template, a ten-item reviewer checklist, a fully worked review of a payment webhook change, size rules for when to demand a split, and the anti-pattern list — buried blockers, comment avalanches, and the silent "later" that never happens.
- It judges the change you show it. A diff pasted without its tests or task context gets an intent pass marked as limited, not a full review.
- It does not run the test suite or execute the code — the verification story is checked by asking what ran and what a regression test would catch, not by running anything.
- Severity defaults assume a production codebase; throwaway prototypes can reasonably relabel some Should fixes as Notes, and the skill says where that judgment applies.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected