1. 1. Summary

Adds an agent skill (`.agents/skills/engflow-profiling/SKILL.md`) documenting how to pull *EngFlow server-side execution profiles*, which are distinct from the Bazel client-side profiles available via the EngFlow MCP. Mirrors 10gen/mms#169467(https://github.com/10gen/mms/pull/169467) into the mongo repo.

1. 1. Why

When investigating RBE wall-clock variance, the Bazel profile shows unexplained silent gaps between "action submitted" and "worker starts executing." Those gaps can be queuing, a retry after OOM/spot reclaim, or a cold CAS fetch — but the Bazel profile can't distinguish them. The EngFlow server-side profile (`/api/profiling/v1/...`) exposes per-action worker assignments, retry attempts (duplicate `action_id` in `worker_time` events), `maxrss` memory usage, action pool assignments, and CAS blob locality (NEAR vs FAR).

1. 1. What's in the skill

• The distinction between Bazel profiles (MCP) and EngFlow profiles (REST API)
• Authentication via the `engflow_auth` credential helper to a Bearer token (`sodalite.cluster.engflow.com`, the same cluster mongo uses)
• The profiling API endpoint format
• Key event types: `worker_time`, `downloadBlob`, `actionCacheLookup`
• Analysis patterns: detecting retries, per-target breakdown, fast-vs-slow comparison
• A decision table for when to use which profile

1. 1. Testing

Documentation only. The example target in the analysis snippet was adapted to a mongo target.