The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses localized shell commands (git and grep) to identify the project structure and list available configuration keys. These operations are limited to metadata discovery (identifying which accounts are configured) and do not expose sensitive values.
[COMMAND_EXECUTION]: The helper script scripts/read-credential.py utilizes subprocess.run with static arguments to find the repository root. This is a standard and safe implementation for path resolution within a development environment.
[DATA_EXPOSURE]: The skill accesses a local .env file to retrieve banking API tokens. This is handled safely by reading the values into environment variables for the CLI tool, avoiding exposure in command-line arguments or process logs. The skill includes explicit instructions for the agent never to print or copy these tokens.
[INDIRECT_PROMPT_INJECTION]: The skill processes financial data retrieved from the Starling Bank API. To mitigate the risk of malicious instructions embedded in transaction data (e.g., in payee names), the instructions explicitly mandate that all retrieved fields must be treated as data and never as instructions.

starling