The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded shared API keys are present in the configuration and root documentation.
Evidence: 9b4e1c73-6a2f-4d88-b5c9-3e7a2f1d6c54 and 2c8f0a91-5d34-4e7b-9a62-f1c3d8e4b705 in src/config/apiKeys.ts and SKILL.md.
[COMMAND_EXECUTION]: The skill exposes Model Context Protocol (MCP) tools that accept raw private keys as parameters to execute direct on-chain transactions, creating a significant security risk for the user's secrets.
Evidence: The execute_cross_perp and execute_isolated_perp tools in mcp-server/src/tools/directExec.ts request privateKey input from the agent.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by fetching and processing untrusted text data from a social comment system without sanitization or boundary markers.
Ingestion points: The getComments function in src/actions/social.ts retrieves user-generated messages.
Boundary markers: Absent; data is returned as raw strings.
Capability inventory: The skill has extensive write permissions, including buyToken and transferNative in src/actions/spotTrade.ts and src/actions/transfers.ts.
Sanitization: No client-side sanitization is performed on comment data before it enters the agent context.

gdex-trading