env per-bench from YAML config; swe image naming to official repo-base

- config env entries: bfcl_v3->bfcl_mock, tau2_bench->tau2_official;
  the eval loop previously POPPED the env key and discarded it -- now
  it feeds args.env (CLI --env still works as default/fallback)
- swe_bench_verified images: per-instance names are never published;
  the official docker.io layout is repo-level BASE images
  (swebench/sweb.eval.x86_64.{repo}) with per-instance images built
  on top -- naming corrected (note: the swebench/* namespace is
  currently blocked on every reachable CN mirror all the same)

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/cli.py

@@ -788,8 +788,12 @@ def _cmd_eval_run(args) -> int:
             # YAML config: per-bench generation params, AUTO-LOADED
             # (single .yaml in config/ = the default; --config overrides)
             bench_cfg = _load_bench_cfg(args, name)
+            # env: per-bench from YAML, CLI --env as fallback/default
+            _env_cfg = bench_cfg.pop('env', '') or ''
+            if _env_cfg:
+                args.env = _env_cfg
             # strip non-generation keys (they go to run_eval kwargs)
-            for k in ('judge', 'judge_url', 'env', 'max_turns',
+            for k in ('judge', 'judge_url', 'max_turns',
                       'limit', 'limit_per_task', 'concurrency'):
                 bench_cfg.pop(k, None)
 

evalharness/config/default.yaml

@@ -54,7 +54,9 @@ openai_mrcr:
   max_input_tokens: 128000
 bfcl_v3:
   max_tokens: 4096
+  env: bfcl_mock          # agent 模式：工具调用轨迹 + 官方 AST 判分
 general_fc:
   max_tokens: 4096
 tau2_bench:
   max_tokens: 16384
+  env: tau2_official      # agent 模式：官方引擎