Loading page
About this tool
Builds an AI code-review prompt focused on correctness, security, performance or style, with severity labels and a review-size check.
The Code Review Prompt Builder assembles an AI code-review prompt from proven checklists — correctness defect classes, OWASP Top 10 (2021) security categories, performance anti-patterns, style and maintainability — with a four-level severity scheme and a review-size check. It is built for developers who paste diffs into ChatGPT, Claude or Copilot Chat and want findings with file, line, failure scenario and a minimal fix instead of a vague rewrite, and it warns when a change exceeds the ~400-line effective review size found in SmartBear's study of peer review at Cisco.
Open Code Review Prompt Builder on AltFTool — it loads instantly in your browser.
Paste your code or data sample into the workspace.
Pick the format, conversion, or analysis you need.
Copy the polished result straight back into your project.
Correctness, security, performance, style and maintainability each carry their own concrete checklist, so the model hunts real defect classes instead of restating the code.
The prompt demands severity, location, a concrete failure scenario and a minimal fix for every finding — and forbids speculative issues without a line to point to.
Changes above about 400 LOC get split into numbered passes, matching the size at which human defect discovery measurably drops.
About 200 to 400 lines per review session. SmartBear's study of peer code review at Cisco found defect-discovery effectiveness drops sharply beyond roughly 400 lines, and recommends reviewing no faster than about 500 lines per hour. This tool computes the number of passes a larger change needs and writes that instruction into the prompt.
Four things: a specific checklist for the chosen focus (for security, the OWASP Top 10 categories such as injection and broken access control), a severity scheme, a required format for each finding (location, defect, failure scenario, minimal fix), and a rule against reporting speculative issues without a concrete line. This tool generates all four from your selections.
No — it is a pre-filter, not a replacement. AI review reliably catches mechanical issues like null handling, injection-prone string building and N+1 queries, but misses context a human has: product intent, architectural direction and team conventions. Use the generated review to arrive at human review with fewer defects.
A four-level scheme is standard: blocker (must fix before merge — bug, vulnerability or data loss), major (should fix — likely defect), minor (worth fixing), and nit (style preference the author may ignore). The generated prompt requires exactly one label per finding so results can be triaged at a glance.