The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is instructed to search for and execute a script named setup-signing.sh from the project root or the user's home directory (~/.config/fastlane/). This pattern allows a repository containing a malicious version of this script to achieve arbitrary code execution on the user's machine when the skill is invoked.
[PROMPT_INJECTION]: The instructions include directives such as "Do NOT ask questions" and "Fully autonomous." These commands attempt to bypass standard agent safety checks and human-in-the-loop confirmation, which are critical for preventing the unintended execution of dangerous operations.
[DATA_EXFILTRATION]: The skill accesses highly sensitive credentials via the AWS CLI (aws secretsmanager get-secret-value). While this is used for deployment, the lack of user confirmation and the presence of arbitrary script execution risk create a direct path for credential exfiltration to an attacker-controlled environment.
[COMMAND_EXECUTION]: The skill performs numerous system-level operations, including modifying project files using sed, querying the system keychain with security find-identity, and executing external build tools like Fastlane and the Flutter CLI.
[DATA_EXFILTRATION]: The skill writes execution metadata and telemetry to a file located at ~/.claude/projects/skill-telemetry.md. Writing to paths outside the immediate project directory for tracking purposes can be used to monitor user activity across different sessions.

flutter-deploy