The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded sensitive data was detected in multiple files within the scripts directory.
File: scripts/archive/check-balances.js contains a plaintext 12-word mnemonic phrase: 'airport room shoe add offer price divide sell make army say celery'.
Similar mnemonics and test private keys are present in other archived scripts including scripts/archive/test-hl-debug.js, scripts/archive/diagnostic-v2.js, and scripts/archive/test-variants.js.
While these appear to be for testing purposes, including active mnemonics in distributed skill files is a significant security risk if users adapt these scripts for production use without replacement.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality.
Ingestion Points: The skill frequently retrieves token metadata, symbols, and names via the getTokenDetails and getTrendingTokens actions from an external API (trade-api.gemach.io).
Capability Inventory: The agent is granted high-privilege capabilities including buy_token, sell_token, and hl_create_order (perpetual futures trading).
Surface: Malicious actors could create tokens with names or symbols containing instructional strings designed to influence agent behavior when the agent 'researches' the token before a trade. The current instructions lack boundary markers or explicit warnings to ignore instructions embedded in token metadata.
[SAFE]: The pre-configured API keys found in src/config/apiKeys.ts are documented as shared keys for public use by AI agents and are considered part of the intended platform architecture.

gdex-trading