From acf0b9b1186b96364e110a86c93a6063b4d13c63 Mon Sep 17 00:00:00 2001 From: Harold Ship Date: Thu, 22 Oct 2026 16:52:21 +0300 Subject: [PATCH] chore(appworld): run AppWorld with execution.enforce_code_security=false - cuga-agent is adding `execution.enforce_code_security` (default true); after that change the `appworld` benchmark profile no longer relaxes code execution on its own - set the flag to false in appworld.env so AppWorld keeps its relaxed mode - harmless before the cuga-agent change lands (Dynaconf carries the unused key) Closes #211 --- benchmarks/appworld/config/appworld.env | 2 ++ 1 file changed, 2 insertions(+) diff --git a/benchmarks/appworld/config/appworld.env b/benchmarks/appworld/config/appworld.env index e8534542..20eb6692 100644 --- a/benchmarks/appworld/config/appworld.env +++ b/benchmarks/appworld/config/appworld.env @@ -12,6 +12,8 @@ DYNACONF_ADVANCED_FEATURES__FORCE_LITE_MODE_APPS=["supervisor","gmail","file_sys EVAL_CONFIG_TOML_PATH="benchmarks/appworld/eval_config.toml" DYNACONF_ADVANCED_FEATURES__DECOMPOSITION_STRATEGY=exact DYNACONF_ADVANCED_FEATURES__BENCHMARK=appworld +# Relaxed execution: the benchmark profile no longer turns off the code security checks; this flag does. +DYNACONF_EXECUTION__ENFORCE_CODE_SECURITY=false DYNACONF_ADVANCED_FEATURES__FORCE_AUTONOMOUS_MODE=true DYNACONF_CONTEXT_SUMMARIZATION__ENABLED=false DYNACONF_ADVANCED_FEATURES__ENABLE_TODOS=false