observability-analyzerlisted

# Observability Analyzer An L2 Truth-Execution skill. Where the `observability` MCP is a **cross-run** tracker (flakiness over time, metric trends, pass rate history), this skill is the **per-run** complement — it opens one run's artifacts (HAR files, Playwright traces, browser console dumps, CDP exports) and produces a report that points at specific anomalies: slow requests, failed dependencies, console errors, performance-budget breaches. Per-run observability is where release decisions catch the "yeah the tests passed but look at the network waterfall" class of regression. The cross-run tool wouldn't see that — it only watches pass/fail + duration. ## When You're Invoked - **PIPELINE-5 step 6** — after `uat-executor` has driven a scenario set against staging and left trace artifacts; the skill reads the traces + surfaces the anomalies before `release-decision-engine` weighs them. - **PIPELINE-6 step 5** — pre-release, same position. - **On demand** as `/vibeflow:observability-analyzer <artifact-path>`. - **From `release-decision-engine`** when a CONDITIONAL verdict needs fresh observability signals before shipping. ## Input Contract | Input | Required | Notes | |-------|----------|-------| | Artifact path | yes | A directory, a file glob, or a single file. Must resolve to at least one supported source. | | `scenario-set.md` | optional but preferred | Needed to map traces to scenarios for the P0 rule. Without it, anomalies still surface but can't be gate-lin