revert eval related changes

Author: jeremymeng

AGENTS.md

@@ -44,11 +44,10 @@ load them all up front.
 
 ## Hard rules
 
-Rules agents have violated in the past and which the Vally eval suites
-under `evals/` (cross-cutting agent-behavior rules live in
-`evals/workflows/mock/`) actively regression-test. Each one is here
-because it has no static-analysis enforcement today (otherwise it would
-be a lint rule or CI check, not an eval).
+Rules agents have violated in the past and which the eval suite at
+`eng/agent-evals/` actively regression-tests. Each one is here because
+it has no static-analysis enforcement today (otherwise it would be a
+lint rule or CI check, not an eval).
 
 - Build a single package with `pnpm turbo build --filter=<pkg>... --token 1`.
   Do **not** use `npm run build` at the package level — turbo workspace
@@ -82,9 +81,8 @@ Rules for the agent that maintains this file:
 
 - A rule that applies to **all** package work → here, "Hard rules"
   section, **only if** it's eval-covered. Don't add hard rules without
-  a corresponding eval under `evals/` (cross-cutting agent-behavior
-  rules go in `evals/workflows/mock/`) — the eval is what prevents the
-  rule from silently drifting.
+  a corresponding eval in `eng/agent-evals/` — the eval is what
+  prevents the rule from silently drifting.
 - A workflow that's specific to one task type (build, generate,
   release, ...) → a new or existing skill in `.github/skills/`.
 - Reviewer-only behavior → `.github/instructions/reviewer/`.