From 8f446567781cc07bfd374f989caf6a0551b424ff Mon Sep 17 00:00:00 2001 From: cris Date: Mon, 29 Sep 2025 08:59:36 +0200 Subject: [PATCH] change env variable CAI_AGENT to CAI_AGENT_TYPE --- benchmarks/Agent2PwnBench/.env.example | 2 +- benchmarks/Agent2PwnBench/README.md | 2 +- benchmarks/Agent2PwnBench/main.py | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/benchmarks/Agent2PwnBench/.env.example b/benchmarks/Agent2PwnBench/.env.example index 76d4aa7e..c9799c21 100644 --- a/benchmarks/Agent2PwnBench/.env.example +++ b/benchmarks/Agent2PwnBench/.env.example @@ -7,6 +7,6 @@ ANTHROPIC_API_KEY="" # The api key of your ANTHROPIC account (use for LLMs in a OLLAMA="" # The api key of your OLLAMA account (use for LLMs in OLLAMA) CAI_STREAM=false # Whether to stream the output of the LLM in CAI or not CAI_MODEL='openai/gpt-4o' # Name of the LLM used in CAI -CAI_AGENT="bug_bounter_agent" # Name of the CAI agent. Available options are (blueteam_agent, web_bounter_agent, redteam_agent ) +CAI_AGENT_TYPE="bug_bounter_agent" # Name of the CAI agent. Available options are (blueteam_agent, web_bounter_agent, redteam_agent ) BURPSUITE_SERVER_URL="http://127.0.0.1:9876/sse" # Local url of the MCP server for burpsuite #for more examples see https://github.com/aliasrobotics/cai/blob/main/.env.example \ No newline at end of file diff --git a/benchmarks/Agent2PwnBench/README.md b/benchmarks/Agent2PwnBench/README.md index 9cc4cd3b..e6cc4e31 100644 --- a/benchmarks/Agent2PwnBench/README.md +++ b/benchmarks/Agent2PwnBench/README.md @@ -46,7 +46,7 @@ The program follows a sequence of steps to evaluate the models. PORTSWIGGER_SECTION='sql-injection' # Change this to the type of lab NUMBER_OF_LABS=10 # To test all the labs in the section, change this to -1 CAI_MODEL='openai/gpt-4o' # LLMs used in CAI - CAI_AGENT="bug_bounter_agent" # Name of the CAI agent. Available options are (blueteam_agent, bug_bounter_agent, redteam_agent) + CAI_AGENT_TYPE="bug_bounter_agent" # Name of the CAI agent. Available options are (blueteam_agent, bug_bounter_agent, redteam_agent) BURPSUITE_SERVER_URL="http://127.0.0.1:9876/sse" # Local url of the MCP server for burpsuite ``` To see more available labs sections, see [**topic_prefixes.json**](utils/topics_prefixes.json) file. diff --git a/benchmarks/Agent2PwnBench/main.py b/benchmarks/Agent2PwnBench/main.py index a686e9f1..5fa8eeba 100644 --- a/benchmarks/Agent2PwnBench/main.py +++ b/benchmarks/Agent2PwnBench/main.py @@ -31,7 +31,7 @@ MODEL = os.getenv("CAI_MODEL") SERVER_URL = os.getenv("BURPSUITE_SERVER_URL") SECTION = os.getenv("PORTSWIGGER_SECTION") N_LABS = int(os.getenv("NUMBER_OF_LABS")) -AGENT = os.getenv("CAI_AGENT") +AGENT = os.getenv("CAI_AGENT_TYPE") REMOVE_LABS_WITH_EXPLOIT_SERVER = False # Set to True to remove labs in portswigger that require an additional exploit server to be solved