The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a shell tool (sample/src/tools/shell.ts) and optional shell shortcut integrations (references/modules.md) that allow for arbitrary command execution. These features are standard for coding agent harnesses but represent a significant capability tier that must be managed carefully by the end developer.
[DYNAMIC_EXECUTION]: The skill includes a js_repl tool (references/tools.md) and uses child_process for shell execution, enabling the runtime execution of code and commands within the generated agent project.
[INDIRECT_PROMPT_INJECTION]: The generated agent infrastructure possesses an attack surface for indirect prompt injection.
Ingestion points: Data enters the agent's context via file_read (sample/src/tools/file-read.ts), grep (sample/src/tools/grep.ts), and web_fetch (references/tools.md).
Boundary markers: The default system prompt generated in sample/src/config.ts does not include explicit delimiters or instructions for the agent to ignore potentially malicious instructions embedded in external files or web content.
Capability inventory: The agent harness includes high-privilege capabilities such as shell, file_write, file_edit, and sub_agent spawning across various tools and optional modules.
Sanitization: There is no documented implementation of prompt sanitization or escaping for content retrieved from external sources before it is interpolated into the conversation history.
[EXTERNAL_DOWNLOADS]: The skill fetches a list of available models from the OpenRouter API via the /model command in sample/src/commands.ts. This is a standard and expected feature for model selection within the vendor's ecosystem.

create-agent-tui