The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to 'Do NOT ask the user questions' and 'Follow the instructions... exactly', which are designed to bypass the normal interactive nature of the agent. This increases the risk of the agent performing sensitive operations autonomously without user confirmation.
[PROMPT_INJECTION]: The skill processes untrusted input through the $ARGUMENTS variable, creating a surface for indirect prompt injection that could influence the generated configuration files and commands.
Ingestion points: $ARGUMENTS input in SKILL.md.
Boundary markers: Absent; user input is directly used to define publishing requirements.
Capability inventory: File system writes (CI/CD workflows, Fastlane configurations) and shell operations (git commits, build execution).
Sanitization: Absent.
[COMMAND_EXECUTION]: The instructions mandate the automated creation and modification of project files, as well as repository commits and build verification steps. This involves the execution of various system commands to configure the development environment and CI/CD pipelines.
[CREDENTIALS_UNSAFE]: The skill explicitly directs the agent to 'Record all required secrets and environment variables' related to code signing and app store access. This instruction could lead to the exposure of sensitive credentials in the agent's output or telemetry logs if they are not handled with strict confidentiality.
[COMMAND_EXECUTION]: The 'Self-Evolution Telemetry' section implements a mechanism to write execution logs to ~/.claude/projects/skill-telemetry.md. This constitutes an automated tracking mechanism that writes to a hidden directory in the user's home folder without explicit user prompt for this specific logging action.

mobile-publish