The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's safety mechanisms and user oversight. Directives such as "Execute them directly without pausing to ask for permission or confirmation" and "Do NOT pause or ask 'Should I run this command?'" are designed to bypass standard agent protection protocols.
[COMMAND_EXECUTION]: The skill authorizes the autonomous execution of shell commands, specifically leveraging the Bun runtime. It mandates that commands like bun run and binary executions should be treated as "pre-approved," removing the human-in-the-loop requirement for sensitive system operations.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of external code from remote registries without manual review. By pre-approving bun install, bun add, and bunx, it allows the agent to download and run arbitrary third-party packages autonomously.
[DYNAMIC_EXECUTION]: The instructions explicitly promote the use of bun -e (eval) for inline code execution. This allows the agent to generate and execute arbitrary scripts at runtime without user verification or confirmation.
[INDIRECT_PROMPT_INJECTION]: The skill creates an exploitable surface for indirect prompt injection.
Ingestion points: The agent interacts with project-controlled files including package.json, scripts, and TypeScript/JavaScript files.
Boundary markers: None. The skill does not define delimiters or provide instructions to ignore embedded commands in project data.
Capability inventory: Full shell access via bun, remote package installation via bunx/install, and arbitrary code evaluation via bun -e.
Sanitization: None. The skill instructs the agent to execute commands immediately based on project context without sanitizing or validating the input.

allow-bun