Catches the mistakes coding agents repeat before they merge: a pull-request gate with no model call and no API key.
Short answer: No eval or new Function on strings; dangerouslySetInnerHTML only with sanitised HTML. Code or HTML built from a string runs whatever the string holds; with any outside input in it, that is an injection.
Group: Security and tenancy. Status: stable. Generated from the rule definitions; do not edit by hand.
Playground: https://brevio.pro/tools/agent-guard-playground?rule=R2.8
No unsafe code or HTML sinks: no eval or new Function on strings; dangerouslySetInnerHTML only with sanitised HTML.
Code or HTML built from a string runs whatever the string holds; with any outside input in it, that is an injection.
eval(userInput);
const value = JSON.parse(userInput);
brevio-core scan: a scan of added lines (code_rules.scan_added_lines_for_rules).A person can silence one finding with brevio-core: ignore R2.8 -- reason on the line or the line above. Without the reason it does not count. An agent cannot silence it: the agent hook and the MCP tools never honour the comment, and a pull request that adds one is reported until a reviewer allows it. Fixing the code is still the better answer.
What the scanner already leaves alone: Comment lines and Playwright’s $eval and $$eval are not flagged; dangerouslySetInnerHTML is checked only in .tsx and .jsx files outside tests.
A scan reads only the lines you add, so a line that was already in the file is not flagged.
To change what the rule flags, edit its definition in dev_rules.py or code_rules/__init__.py. That is a code change and is reviewed like one.