The Agent Skills Directory

[SAFE]: The skill implements a legitimate business application for contract management. All functionality is consistent with its stated purpose and documented boundary.
[DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration or credential exposure. The application reads and writes data exclusively to the Busabase platform or the local demo environment as intended. Deployment scripts correctly utilize environment variables for sensitive tokens.
[COMMAND_EXECUTION]: Shell commands are restricted to standard development and deployment tasks (e.g., pnpm dev, publish_airapp.mjs) which are explicitly documented for the operator and do not process unsanitized user input.
[REMOTE_CODE_EXECUTION]: The skill does not perform any unauthorized remote code execution. Dependencies are pinned to specific versions and originate from recognized registries.
[INDIRECT_PROMPT_INJECTION]: The skill handles contract data which could potentially contain malicious instructions, but it mitigates this risk through proper architecture.
Ingestion points: Data is fetched from Busabase via busabaseProvider.getState() in content/kelly-clm-app/app/js/providers/busabase-provider.js.
Boundary markers: The instructions explicitly define the tool as an operations tracker and warn against legal or contractual side effects.
Capability inventory: Capabilities are restricted to Busabase record updates; no direct shell or general network access is provided to the runtime logic.
Sanitization: Content is escaped in app.js using the e(value) helper before rendering.
[DYNAMIC_EXECUTION]: Build-time bundling using esbuild-wasm is used for preparing vendor scripts, which is a standard development practice and does not occur at runtime based on untrusted input.

kelly-clm