The Agent Skills Directory

[SAFE]: The skill is authored by 'helius-labs' and utilizes official vendor resources, including 'helius-rpc.com', 'helius.xyz', and 'dflow.net'. All external references are consistent with the primary purpose of building Solana trading tools.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: While the skill references sensitive configuration paths like '~~/.config/dflow/config.json', '~~/.ows/', and '~/.helius/keypair.json', it does so in the context of documenting legitimate wallet management tools. The instructions explicitly recommend using environment variables for API keys and emphasize that private keys should never be embedded in code or logs. The 'dflow wallet export' command is a documented user-initiated action for local wallet management.\n- [REMOTE_CODE_EXECUTION]: The skill adopts a strong security posture by instructing the agent not to fetch or execute the DFlow Agent CLI installation script on the user's behalf, instead directing the user to the official documentation to perform the installation manually. This prevents unauthorized execution of remote code.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface involving wallet addresses, market tickers, and search queries. Evidence: 1. Ingestion points include user-supplied addresses and market data in SKILL.md and dflow-prediction-markets.md. 2. Capability inventory includes transaction submission and trade execution. 3. Sanitization is addressed in dflow-prediction-markets.md, which states that special characters are escaped before search. The risk is mitigated by the transactional nature of the operations and the requirement for user confirmation or pre-configured guardrails.\n- [COMMAND_EXECUTION]: The skill uses the DFlow Agent CLI for trading operations. These commands are executed locally using the user's own infrastructure and are protected by HMAC-signed guardrails that prevent agents from modifying safety limits.

dflow