The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to run npx lazycodex-ai install and npx --package oh-my-openagent omo install. These commands download and execute code from external, third-party packages that are not verified or from a trusted source.
[EXTERNAL_DOWNLOADS]: The skill depends on the oh-my-openagent package, which is hosted at an external GitHub repository (github.com/code-yeongyu/oh-my-openagent) and the lazycodex-ai package from the NPM registry.
[COMMAND_EXECUTION]: The skill implements powerful autonomous coding commands such as $ulw-loop and $start-work. The $ulw-loop command runs a self-referential execution loop that can iterate up to 500 times, performing automated file modifications and running shell commands without mandatory human review between iterations.
[DYNAMIC_EXECUTION]: The skill documentation describes defining and loading custom functionality through the defineSkill and defineHook APIs. This indicates the framework dynamically loads code modules from user-defined paths into the execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through task descriptions and codebase content.
Ingestion points: User-provided task descriptions passed to the $ulw-loop command and the contents of AGENTS.md or other source code files processed during autonomous coding tasks in SKILL.md.
Boundary markers: The instructions do not define explicit delimiters or warnings to ignore embedded commands within the processed codebase data.
Capability inventory: The skill possesses extensive capabilities across all files including shell command execution via npx, multi-agent coordination, and project-wide file write access.
Sanitization: There is no mention of sanitization or validation logic for external task strings or the content of files read by the project memory system.

lazycodex-agent-harness