The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill collects project metadata and constraints from the user and writes them into the CLAUDE.md file, which provides persistent context for the agent's future behavior in the project.
Ingestion points: User-provided answers for project name, description, business model, and constraints (Step 3).
Boundary markers: None; user input is interpolated directly into the Markdown template.
Capability inventory: The agent can execute shell commands, read/write files, and manage dependencies.
Sanitization: None; instructions mandate using the user's exact wording, which could allow for embedded instructions.
[DYNAMIC_EXECUTION]: Employs node -e one-liners to perform search-and-replace operations in the .env file. This involves concatenating user-provided values into a command-line script, which is a potential surface for command injection if the input contains shell-sensitive characters.
[COMMAND_EXECUTION]: Executes standard initialization commands including npm install for dependency management and npm run db:migrate for database schema setup.

init-project