The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to handle sensitive blockchain private keys through the DEFENDER_PRIVATE_KEY environment variable. This key is used in scripts/ids_engine.py and scripts/defense_executor.py to sign and broadcast transactions to the Ethereum network. Exposure of this key would grant full control over the associated account.
[COMMAND_EXECUTION]: The scripts/exploit_classifier.py script implements a load_model function that uses pickle.load() to deserialize machine learning models from a file. This is a known unsafe operation that can lead to arbitrary code execution if an attacker provides a compromised model file.
[EXTERNAL_DOWNLOADS]: The skill establishes outbound network connections via WebSockets and HTTP to blockchain infrastructure providers (such as Alchemy or Infura) and can trigger alerts via external webhooks as configured in scripts/ids_engine.py.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transaction data from the Ethereum mempool to make security decisions.
Ingestion points: scripts/mempool_monitor.py retrieves pending transactions from external blockchain nodes.
Boundary markers: None identified; raw hex data and transaction metadata are ingested directly.
Capability inventory: The skill has the capability to sign and send transactions (scripts/defense_executor.py), effectively spending funds and interacting with smart contract states.
Sanitization: The scripts/payload_analyzer.py performs heuristic and entropy-based analysis, but the system relies on machine learning classification which is inherently susceptible to adversarial inputs.

Mempool Intrusion Detection System