The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local project documentation and configuration files to determine the setup flow and required credentials. This creates an attack surface for indirect prompt injection if those files contain malicious or misleading instructions designed to influence the AI's generation of the setup script.
Ingestion points: Processes files such as README, .env.example, and framework configurations within the local repository as specified in the SKILL.md scoping phase.
Boundary markers: The instructions do not define specific delimiters or warnings for the agent to distinguish between its own logic and the untrusted content found in repository files.
Capability inventory: The generated wizards have the capability to write to local files (write_env) and interact with the GitHub CLI (set_secret) to store sensitive data.
Sanitization: The skill lacks explicit instructions for sanitizing or validating information extracted from repository files before incorporating it into the interactive prompts of the generated script.
[COMMAND_EXECUTION]: The generated wizards are designed to execute shell commands to perform configuration actions.
Evidence: The provided template.sh includes helper functions that execute system utilities such as gh (GitHub CLI) for secret management and platform-specific commands like xdg-open or explorer.exe for opening URLs.
Context: While these executions are aligned with the skill's purpose, they provide the mechanism through which the wizard interacts with the user's environment and external services.

wizard