The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/preflight.py and scripts/gateway.py files use subprocess.run to call the local pear_mcp.py script. This is a legitimate architectural choice to modularize authentication and tool calling within the skill's own file structure. The commands are constructed using static paths relative to the script location (os.path.join(_here, 'pear_mcp.py')) and use the current interpreter (sys.executable), which prevents typical command injection vectors.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with several external APIs including pro-gateway.pearprotocol.io, mcp.pearprotocol.io, and api.hyperliquid.xyz. These are well-known services associated with the skill's primary purpose (Pear Protocol and Hyperliquid). The automated scanner flagged api.hyperliquid.xyz for remote code execution; however, the code actually performs standard JSON-RPC queries (clearinghouseState, maxBuilderFee) and parses the resulting data. This is standard DeFi integration behavior and not an execution of remote code.\n- [CREDENTIALS_UNSAFE]: The skill manages sensitive data including PEAR_API_KEY, Bearer tokens, and Hyperliquid signer keys. It follows security best practices by storing tokens in a restricted directory (/data/workspace/.pear) with 0600 file permissions and recommending the use of .env files for API keys. It does not contain hardcoded secrets; it uses placeholders like YOUR_API_KEY_HERE in documentation and dynamically generates keys for the user during the setup flow.\n- [DATA_EXFILTRATION]: The skill reads account equity and wallet addresses. This data is exclusively sent to the official Pear Protocol gateway and Hyperliquid exchange for the purpose of trade execution and account management. There are no secondary network destinations or suspicious data harvesting patterns.

pear-protocol