The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions designed to suppress human-in-the-loop verification, specifically 'You are in AUTONOMOUS MODE. Do NOT ask questions' and 'Execute the full pipeline below without pausing for user input'. These directives attempt to override the agent's default safety behavior of seeking user confirmation for sensitive operations.
[COMMAND_EXECUTION]: The skill performs extensive shell command execution for environment configuration and dependency management using package managers such as npm, pip, and flutter, as well as platform-specific tools like pod and npx.
[EXTERNAL_DOWNLOADS]: The skill initiates the installation of multiple third-party libraries and SDKs from official registries and fetches service worker scripts from Google's gstatic content delivery network.
[CREDENTIALS_UNSAFE]: The skill automates the configuration and storage of highly sensitive platform credentials, including Firebase service account private keys and Apple Push Notification service (APNs) Auth Keys (.p8 files). It instructs the agent to store these secrets in local project environment files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting project metadata from files like 'package.json' and 'pubspec.yaml'. Evidence: (1) Ingestion: Scans configuration files for platform detection. (2) Boundary markers: No delimiters or ignore instructions are used for the ingested content. (3) Capability inventory: The skill has file-write and subprocess execution capabilities. (4) Sanitization: No input validation is performed on the data read from project files.

push-notifications