The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of various hithink-finance CLI commands to manage the environment, including auth, config, doctor, update, uninstall, and skills management.
[CREDENTIALS_UNSAFE]: Instructions direct the agent to access and read API keys from hithink-finance/credentials.env located in sensitive system directories such as %APPDATA% (Windows), ~/Library/Application Support/ (macOS), and ~/.config (Linux).
Evidence: The agent is tasked with injecting these credentials into the process environment or via stdin for authentication.
Mitigation: The skill includes a 'Boundary Statement' explicitly forbidding the agent from repeating the API key or writing it to logs, Git, or Markdown files.
[EXTERNAL_DOWNLOADS]: The hithink-finance update command performs network operations to check for and install the latest versions of the CLI tool from the npm registry.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of CLI command outputs and manages the synchronization of other agent skills, creating a surface for indirect injection if upstream data or skill manifests are compromised.
Ingestion points: CLI command output (JSON format) and credentials.env file content.
Boundary markers: Relies on structured JSON envelopes and explicit instruction-based constraints for the agent.
Capability inventory: Shell command execution, file system modification (linking/copying skills), and network access for updates.
Sanitization: Employs JSON formatting for machine reading and instructions to summarize large outputs rather than echoing them verbatim.

hithink-finance-shared