Output Modes

Control console verbosity and prompt language with OUTPUT_MODE and PROMPT_LANGUAGE.

LuaN1aoAgent provides three console output modes and a prompt language switch. These settings affect what you see in the terminal during a run — they do not change the agent’s behaviour or the content of log files.

---

Setting the output mode

Environment variable

.env

# "simple", "default", or "debug"
OUTPUT_MODE=default

CLI flag

python agent.py --output-mode debug --target http://example.com

The --output-mode flag overrides the environment variable for that run.

---

Mode comparison

	simple	default	debug
Final result / flag	Yes	Yes	Yes
Task plan summary	No	Yes	Yes
Per-step tool calls	No	Yes	Yes
Tool outputs (truncated)	No	Yes	Yes
Reflector analysis	No	No	Yes
Causal graph updates	No	No	Yes
LLM request / response payloads	No	No	Yes
Token usage per call	No	No	Yes
Full tool stdout (untruncated)	No	No	Yes

---

simple

Minimal output — only the final result or captured flag is printed. Suitable for automated pipelines, CI environments, or batch runs where you only care about the outcome.

OUTPUT_MODE=simple python agent.py --target http://192.168.1.10

Example output:

[+] Mission complete
Flag: CTF{3xf1ltr4t3d_s3cr3t}

---

default

Standard output for interactive use. Shows the current task plan, each executor step, tool names and arguments, truncated tool outputs, and a brief summary at the end of each P-E-R cycle. This is the recommended mode for most users.

OUTPUT_MODE=default python agent.py --target http://192.168.1.10

Example output:

[Planner] Building task graph...
[Task 1] Enumerate web directories
  [Step 1] dirsearch_scan(url="http://192.168.1.10", extensions="php,html")
  [Output]  200 /admin/login.php
           200 /backup/db.sql
[Reflector] Updating causal graph...
[+] Cycle 1 complete. 2 new artifacts.

---

debug

Verbose output equivalent to --verbose. Prints everything default shows plus full LLM request/response payloads, token usage per API call, reflector causal graph diffs, and untruncated tool output. Use this when diagnosing unexpected agent behaviour.

OUTPUT_MODE=debug python agent.py --target http://192.168.1.10

Debug mode can produce very large amounts of output for long runs. Consider redirecting to a file:

OUTPUT_MODE=debug python agent.py --target http://192.168.1.10 2>&1 | tee run.log

---

Prompt language

PROMPT_LANGUAGE controls the language used in all internal agent prompts — system messages, planning instructions, and reflector directives sent to the LLM.

.env

# "zh" (Chinese, default) or "en" (English)
PROMPT_LANGUAGE=en

Value	Description
zh	Chinese prompts (default). Used in the original research and benchmarks.
en	English prompts.

PROMPT_LANGUAGE affects the language of the prompts sent to the LLM, not the language of console output or log files. Tool output and LLM responses will still appear in whatever language the model produces.

If you are using a model that performs better in English (e.g., GPT-4o, Claude), set PROMPT_LANGUAGE=en. For models fine-tuned on Chinese data (e.g., some DeepSeek variants), the default zh may yield better results.

---

Log files

Console output mode does not affect file logging. Logs are always written at full verbosity to the logs/ directory:

File	Contents
logs/mcp_service.log	MCP server tool execution events
logs/agent_*.log	Per-run agent trace (created at run start)

To adjust the log level for file output, set the LOG_LEVEL environment variable:

.env

# DEBUG, INFO, WARNING, ERROR
LOG_LEVEL=INFO