info/                Sample metadata for the benign, MalRadar, and new malware splits.
model_registry.yaml Model/provider configuration for LLM-based stages.
DATA.md              Dataset composition, provenance, and access instructions.
LICENSE.txt          License scope and third-party-material exclusions.
CITATION.cff         Machine-readable citation metadata.
src/                 Static analysis, summarization, behavior generation, and metrics code.

MalEval/
|-- info/
|-- src/
`-- artifacts/
    |-- apk/
    |-- call_chain/
    |-- entrypoints/
    |-- functions/
    |-- meta_info/
    |-- reachable_func/
    `-- reports/

python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install --upgrade pip
python3 -m pip install -r requirements.txt

models:
  gpt:
    provider: openai
    model: gpt-5
    api_key: <your_openai_key>
    base_url: https://api.openai.com/v1

sha=<apk_sha256>
python3 src/preparation/run_static_analysis.py \
  --folder malradar \
  --apk "$sha" \
  --output-root results/static_analysis

python3 src/preparation/run_static_analysis.py \
  --folder malradar \
  --sample-size 5 \
  --seed 42 \
  --output-root results/static_analysis

results/static_analysis/entrypoints/
results/static_analysis/call_chain/
results/static_analysis/functions/
results/static_analysis/reachable_func/

model=gpt
split=malradar
sha=<apk_sha256>

python3 src/preparation/get_functionality_summary.py --model "$model" --folder "$split" --apk "$sha"

python3 src/preparation/get_context_summary.py --model "$model" --folder "$split" --apk "$sha"
python3 src/preparation/get_no_context_summary.py --model "$model" --folder "$split" --apk "$sha"

python3 src/generate_behavior/get_behavior.py --model "$model" --folder "$split" --flag context --apk "$sha"
python3 src/generate_behavior/get_behavior.py --model "$model" --folder "$split" --flag no_context --apk "$sha"
python3 src/generate_behavior/get_meta_behavior.py --model "$model" --folder "$split" --apk "$sha"

python3 src/metrics/run_metrics.py --model "$model" --flag context
python3 src/metrics/run_metrics.py --model "$model" --flag no_context
python3 src/metrics/run_metrics.py --model "$model" --flag meta

python3 src/metrics/run_metrics.py --model "$model" --flag context --skip-eas

python3 src/metrics/aec_b_f1.py --model "$model" --flag context
python3 src/metrics/fpcr_tpmr_f1c.py --model "$model" --flag context
python3 src/metrics/eas.py --model "$model" --flag context --split malradar --judge-model gpt-5

python3 -m compileall -q src
python3 - <<'PY'
import json
from pathlib import Path

expected = {
    "archived_sample_info.json": 200,
    "latest_sample_info.json": 30,
    "benign_sample_info.json": 25,
}
for name, count in expected.items():
    actual = len(json.loads((Path("info") / name).read_text()))
    assert actual == count, (name, actual, count)
print("metadata check passed: 200 archived malware + 30 recent malware + 25 benign")
PY