The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute significant system-level commands to scaffold, build, and manage applications.
Evidence includes instructions to run laravel new, npx create-next-app, php artisan migrate, and start local development servers during the 'Vibe coding studio' phase.
It performs Git operations including commits and codebase mapping.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its architecture of ingesting untrusted content while holding high-privilege capabilities.
Ingestion points: Reads data from external sources like Reddit, Product Hunt, and Indie Hackers via web search. It also reads local files when onboarding existing projects.
Boundary markers: The instructions do not define strict boundary markers or 'ignore' instructions for data retrieved from web searches or local files.
Capability inventory: The agent can write files, execute shell commands (framework CLIs, package managers), and perform network operations (web search).
Sanitization: There is no mention of sanitizing or escaping content retrieved from external sources before it is processed by the subagents or the main thread.
[DATA_EXFALTRATION]: The skill explicitly instructs the agent to scan existing projects for sensitive architectural components.
In 'Existing project mode', the agent is told to scan for 'auth, payments, DB, tests, token tracking, memory'.
While the stated purpose is for auditing and management, this pattern involves the automated discovery of sensitive application logic which could be misused if the agent is compromised via indirect injection.

idea-validation