The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests financial transaction data, such as descriptions and counterparty names, from external providers into a centralized database. While the agent is instructed to summarize this data, there is a theoretical surface for indirect prompt injection if transaction metadata contains malicious instructions. However, the skill implements strong mitigations, including a strictly read-only frontend and a sync process that requires explicit user approval for any mutation.
[SAFE]: The skill adheres to credential safety standards by explicitly instructing the agent never to ask for secrets in chat or store them in public repositories. It directs users to manage API keys via local environment files.
[SAFE]: The project includes comprehensive linting and verification scripts that enforce deployment standards and visual consistency, ensuring the skill operates within its intended Cloud-only or local-preview boundaries.
[SAFE]: All external dependencies, such as the Busabase SDK and Hono, originate from established and trusted sources, and the skill's own code does not perform unauthorized network operations or dynamic code execution.

kelly-money