The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts including scripts/mcp-oauth-refresh.sh for managing authentication tokens and ./notify for reporting account status. It also calls MCP-based tools to execute financial orders.
[INDIRECT_PROMPT_INJECTION]: Financial trades are triggered by the ${var} input variable. If this variable is populated from untrusted sources, an attacker could potentially inject a command to place unauthorized orders.
Ingestion points: The ${var} variable is the primary controller for switching between read-only reporting and active trading.
Boundary markers: No explicit delimiters or sanitization protocols are mentioned to separate instructions from data within the variable input.
Capability inventory: The skill possesses the capability to execute market and notional orders through the mcp__robinhood-trading__* toolset.
Sanitization: Although the skill performs sanity checks on the trade parameters (symbol, size, and available buying power), it does not verify the original source or intent of the instruction.
[DATA_EXFILTRATION]: The skill accesses highly sensitive brokerage data, such as portfolio value, specific asset positions, and order history. It also utilizes high-privilege credentials, including GH_SECRETS_PAT and OAuth tokens, to facilitate account access.

robinhood-mcp