A multi-line comment block that walks through what the function does in prose. The signature already says it; the prose is decorative.
Each example shows the exact code shape the rule is looking for, then the smallest version that keeps the intent without hiding risk or adding noise.
// This function takes a user ID and a list of orders,
// filters the orders to only those belonging to the user,
// then sums the total amount across them and returns it.
function sumUserOrders(userId: string, orders: Order[]): number {
return orders
.filter(o => o.userId === userId)
.reduce((sum, o) => sum + o.amount, 0)
}Deterministic comment-block parse: flags consecutive comment lines that paraphrase the immediately following declaration.
function sumUserOrders(userId: string, orders: Order[]): number {
return orders
.filter(o => o.userId === userId)
.reduce((sum, o) => sum + o.amount, 0)
}Delete the prose that repeats the code and leave the implementation readable through names and structure. Keep a comment only when it explains intent, a constraint, or an external contract the code cannot express.
Agents are trained to "explain" code. They pad function bodies with explanatory preambles even when the signature, parameter names, and operator chain already tell the same story.
ai-slop/narrative-comment
ai-slop
Deterministic comment-block parse: flags consecutive comment lines that paraphrase the immediately following declaration.
WHY comments, license headers, doc comments documenting public API contracts, and comments stating a non-obvious constraint are not flagged.
Verbosity signals derive from SlopCodeBench, which measured comment padding in long-horizon coding tasks, and recurred across the first public scan batch.