The Agent Skills Directory

[PROMPT_INJECTION]: No attempts to override agent behavior or bypass safety filters were found. The skill includes legitimate instructions for output language (Korean) and architectural enforcement without attempting to extract system prompts or disregard safety guidelines.
[DATA_EXFILTRATION]: No evidence of hardcoded credentials or unauthorized data transmission. The skill explicitly contains 'Blocking Safety Gates' and 'Security Rules' designed to prevent the accidental exposure of secrets, database clients, or privileged SDKs to the client-side code.
[REMOTE_CODE_EXECUTION]: The skill uses standard development tools and commands (ls, grep, ripgrep) for project validation and auditing. It references official TanStack packages and does not download or execute scripts from untrusted remote sources.
[COMMAND_EXECUTION]: The shell commands used in the validation sections (e.g., rg, find, wc) are standard for source code analysis and are used within the scope of the skill's purpose to audit project structure and safety.
[CREDENTIALS_UNSAFE]: No hardcoded secrets or API keys. The skill includes rules to ensure that environment variables like DATABASE_URL and SECRET are managed securely using tools like @t3-oss/env-core and are never exposed to the client.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests and processes project files (source code, config), it does so to perform architectural reviews. The 'Blocking Safety Summary' provides guardrails against the agent performing unsafe actions based on the content of the analyzed files.
[SAFE]: The skill is highly focused on improving the security posture of TanStack Start applications by enforcing import boundaries, validating server function inputs, and ensuring hydration stability.

tanstack-start-architecture