trade-flap-token
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes robust defenses against malicious data embedded in external sources like token names or symbols.
- Ingestion points: The skill ingests token addresses from user input and reads token metadata (names, symbols) from the blockchain via
references/token-state.md. - Boundary markers:
SKILL.mdandreferences/safety.mdcontain explicit directives to ignore any instructions found in token metadata, descriptions, or forwarded messages. - Capability inventory: The skill facilitates transaction execution through the
bawCLI tool and local simulations. - Sanitization: The instructions mandate that all data retrieved from the blockchain be treated as data only, never as instructions for the agent.
- [DYNAMIC_EXECUTION]: To ensure functionality across different environments, the skill provides a fallback mechanism in
references/preflight.mdwhere the agent can generate a temporary script to perform EVM operations. This script uses the well-known and trustedviemlibrary and is executed in a transient directory, representing a controlled use of dynamic code generation. - [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the
@binance/agentic-walletpackage from the npm registry and connects to the official Binance RPC endpoint. These are recognized as legitimate vendor resources and well-known services. The skill properly requires user consent before any software is installed on the host system. - [COMMAND_EXECUTION]: The skill relies on the
baw(Binance Agentic Wallet) CLI tool to perform its primary functions, including wallet status checks, transaction previews, and trade execution. All critical operations (transaction sending) are gated by a mandatory user confirmation step. - [PROMPT_INJECTION]: Instructions in
SKILL.mdguide the agent to batch background processing tasks to improve performance and user experience. While this involves minimizing intermediate dialogue about sub-steps, it is a benign UX optimization. The skill maintains security integrity by requiring an explicit and standalone confirmation from the user before any funds or allowances are moved.
Audit Metadata