refactor: move live-model benchmark harnesses from scripts/ into evals/

scripts/ is for repo tooling (tests runner, release, installers, CI
checks). The tool-search live tests, the toolperf A/B eval and the browser
eval benchmark are offline benchmarks that spend model budget, which is
exactly what evals/ holds; evals/browser_use already cited
scripts/toolperf_abeval as "the same pattern".

- scripts/tool_search_livetest*.py + analyze_livetest.py + LIVETEST_README
  -> evals/tool_search/ (README.md); repo-root sys.path hop adjusted for
  the extra directory level; gitignore now covers evals/tool_search/out*/
- scripts/toolperf_abeval/ -> evals/toolperf_abeval/
- scripts/benchmark_browser_eval.py -> evals/browser_use/

.gitignore

@@ -190,8 +190,8 @@ docs/superpowers/*
 /.install_method
 
 # Tool Search live-test harness output — non-deterministic model transcripts,
-# regenerated by scripts/tool_search_livetest.py. Never an artifact of the repo.
+# regenerated by evals/tool_search/tool_search_livetest*.py. Never an artifact of the repo.
-scripts/out/
+evals/tool_search/out*/
 
 # Per-release changelog drafts. These exist only transiently during a release
 # cut (passed to `gh release create --notes-file`); the GitHub Release itself

evals/browser_use/README.md

@@ -21,7 +21,7 @@ web tasks.
   rating aggregation, JS/delayed render, login chain, cross-category
   compare).
 - **Resume-safe.** Completed cells in `results/*.jsonl` are skipped on rerun
-  (same pattern as `scripts/toolperf_abeval`).
+  (same pattern as `evals/toolperf_abeval`).
 - **Backend matrix.** `orchestrate.py` drives a local headless-Chrome CDP;
   `orchestrate_cloud.py --backend nous-cloud|browserbase` provisions a real
   cloud browser per cell through the same provider plumbing the product uses.

evals/browser_use/benchmark_browser_eval.py

@@ -4,7 +4,7 @@ Runs both paths against the same live Chrome and prints a comparison table.
 Not a pytest — a script you run manually for the PR description.
 
 Usage:
-    .venv/bin/python scripts/benchmark_browser_eval.py [--iterations N]
+    .venv/bin/python evals/browser_use/benchmark_browser_eval.py [--iterations N]
 """
 from __future__ import annotations
 

evals/browser_use/orchestrate.py

@@ -1,7 +1,7 @@
 """Local-CDP battery orchestrator: tasks x arms x models x reps.
 
 Resume-safe: completed cells in results.jsonl are skipped, so a killed
-battery continues where it left off (same pattern as scripts/toolperf_abeval).
+battery continues where it left off (same pattern as evals/toolperf_abeval).
 
 Usage:
     # start a headless Chrome first:

evals/tool_search/README.md

@@ -2,14 +2,14 @@
 
 Runs five scenarios against a real model (Claude Haiku 4.5 via OpenRouter) to
 verify that the bridge tools work end-to-end. Records transcripts in
-`scripts/out/`.
+`evals/tool_search/out/`.
 
 ## Running
 
 ```bash
 cd <repo root>
-python3 scripts/tool_search_livetest.py        # runs all 5 scenarios x 2 modes
+python3 evals/tool_search/tool_search_livetest.py        # runs all 5 scenarios x 2 modes
-python3 scripts/analyze_livetest.py            # side-by-side report
+python3 evals/tool_search/analyze_livetest.py            # side-by-side report
 ```
 
 Requires `OPENROUTER_API_KEY` set or present in `~/.hermes/.env`.
@@ -34,7 +34,7 @@ A/B baseline. The harness records:
 ## Output structure
 
 ```
-scripts/out/
+evals/tool_search/out/
   <scenario>__enabled.json    # tool_search ON
   <scenario>__disabled.json   # tool_search OFF
   _summary.json               # one-line summary across all runs

evals/tool_search/tool_search_livetest.py

@@ -37,7 +37,7 @@ ORIGINAL_HOME = os.environ.get("HERMES_HOME")
 ORIGINAL_AUTH = Path.home() / ".hermes" / "auth.json"
 
 _THIS_DIR = Path(__file__).resolve().parent
-_WORKTREE_ROOT = _THIS_DIR.parent
+_WORKTREE_ROOT = _THIS_DIR.parents[1]
 sys.path.insert(0, str(_WORKTREE_ROOT))
 
 # ---------------------------------------------------------------------------

evals/tool_search/tool_search_livetest2.py

@@ -16,7 +16,7 @@ from pathlib import Path
 from typing import Any, Dict, List
 
 _THIS_DIR = Path(__file__).resolve().parent
-_WORKTREE_ROOT = _THIS_DIR.parent
+_WORKTREE_ROOT = _THIS_DIR.parents[1]
 sys.path.insert(0, str(_WORKTREE_ROOT))
 sys.path.insert(0, str(_THIS_DIR))
 

evals/tool_search/tool_search_livetest_ue.py

@@ -23,7 +23,7 @@ from pathlib import Path
 from typing import Any, Dict, List
 
 _THIS_DIR = Path(__file__).resolve().parent
-_WORKTREE_ROOT = _THIS_DIR.parent
+_WORKTREE_ROOT = _THIS_DIR.parents[1]
 sys.path.insert(0, str(_WORKTREE_ROOT))
 sys.path.insert(0, str(_THIS_DIR))
 

evals/tool_search/tool_search_livetest_ue_disc.py

@@ -24,7 +24,7 @@ from pathlib import Path
 from typing import Any, Dict, List
 
 _THIS_DIR = Path(__file__).resolve().parent
-_WORKTREE_ROOT = _THIS_DIR.parent
+_WORKTREE_ROOT = _THIS_DIR.parents[1]
 sys.path.insert(0, str(_WORKTREE_ROOT))
 sys.path.insert(0, str(_THIS_DIR))
 

evals/tool_search/tool_search_livetest_ue_hard.py

@@ -27,7 +27,7 @@ from pathlib import Path
 from typing import Any, Dict, List
 
 _THIS_DIR = Path(__file__).resolve().parent
-_WORKTREE_ROOT = _THIS_DIR.parent
+_WORKTREE_ROOT = _THIS_DIR.parents[1]
 sys.path.insert(0, str(_WORKTREE_ROOT))
 sys.path.insert(0, str(_THIS_DIR))
 

evals/toolperf_abeval/README.md

@@ -55,7 +55,7 @@ in real production traffic.
 ## Run
 
 ```bash
-cd scripts/toolperf_abeval
+cd evals/toolperf_abeval
 export ABEVAL_ROOT=/tmp/abeval-workspace   # results + sandboxes land here
 export ABEVAL_HOME=/tmp/abeval-home
 ./run_all.sh /tmp/abeval-baseline /path/to/fixes-tree 3 \