The Agent Skills Directory

[SAFE]: The skill follows established architectural patterns for App-in-Skill extensions, maintaining a clear boundary between agent drafting and human approval.
[CREDENTIALS_UNSAFE]: Authentication with the Busabase backend is handled through standard environment variables and ambient sessions. The skill instructions explicitly warn against handling or asking for secret tokens in the chat interface, following best practices for credential safety.
[EXTERNAL_DOWNLOADS]: The skill relies on official packages like busabase-sdk and hono. All dependencies are explicitly pinned to specific versions in package.json to prevent supply chain drift.
[DATA_EXFILTRATION]: Network activity is restricted to communication with the Busabase API for task management. There is no evidence of unauthorized data transmission to external domains.
[COMMAND_EXECUTION]: Utility scripts provided in the scripts/ directory are intended for administrative tasks like workspace setup and app publishing. These are controlled by the user and do not contain malicious payload triggers.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided product documentation. It mitigates indirect prompt injection risks by implementing a strict human-in-the-loop approval workflow where all agent-generated drafts must be reviewed before execution.
[DYNAMIC_EXECUTION]: The sync-content.mjs script uses dynamic imports for loading configuration files during the build process, which is a standard development practice and does not process untrusted external data.

kelly-launch