What problem does it solve? Code often claims behaviors (caching, parallelism, internal routing) that production traffic contradicts, and these runtime-shape defects slip past diff review and unit tests. This Skill diffs what the code assumes against what Datadog traces actually show, turning mismatches into verifiable fix tickets. ## Core Features & Use Cases - Runtime Profile Building: Pulls a Datadog trace via MCP and profiles per-resource call counts, durations, target hosts, and sequential-vs-parallel timelines. - Assumed-vs-Actual Diffing: Extracts declared assumptions from the code (cache annotations, TTLs, async constructs, base URLs) and applies a language-neutral heuristics catalog to detect defects like bypassed caches, N+1 patterns, and wrong-edge routing. - Fix Ticket Generation: Creates tickets following ar-ticket-creator conventions, each embedding a runtime budget in its "Done when" so re-running the audit objectively verifies the fix. - Use Case: An endpoint feels slow in production. You provide a Datadog trace URL; the audit reveals the same currency-rate fetch runs five times per request despite a declared cache, and files a ticket requiring at most one fetch per request. ## Quick Start Say "ar-api-dd-compare" or "trace audit" and provide a Datadog trace URL or endpoint name with a time window to start the audit.
