The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow project-level instructions from files such as AGENTS.md and CLAUDE.md in the target repository. This creates a surface where malicious instructions embedded in a project's documentation could influence the agent's behavior during the app generation or upgrade process.
Ingestion points: Files like package.json, vite.config.ts, AGENTS.md, and CLAUDE.md are read into the context in SKILL.md (Step 2) and by the scripts/audit_app_project.py script.
Boundary markers: The instructions do not define strict delimiters or 'ignore' commands for the content of these files.
Capability inventory: The skill has extensive capabilities including shell command execution (pnpm, python3, tmux), file system writes (project scaffolding), and network operations via curl to local dev servers.
Sanitization: No sanitization or validation of the content of project-level instruction files is performed before the agent is told to 'honor' them.
[COMMAND_EXECUTION]: The skill uses various shell commands to perform its duties, including project initialization (pnpm create vite), dependency management (pnpm add), and environment management (tmux). It also executes a local Python script (scripts/audit_app_project.py) to verify project structure. While these are typical for a development tool, they represent a significant capability tier.

lov-app-generator