The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute a script located at a hardcoded absolute path: bun ~/.config/LIFEOS/runtime/TOOLS/SecretScan.ts .. Running scripts from fixed locations outside the project's scope, particularly within a user's configuration directory, is a high-risk operation as the script content is unverifiable and could execute arbitrary malicious code.
[DYNAMIC_EXECUTION]: The 'Skill Extraction' feature (scripts/extract-skill.ts) is designed to dynamically generate new SKILL.md files (agent instructions), README.md, and openai.yaml files. This capability to create new agent logic at runtime can be exploited to generate malicious skills that bypass safety constraints or automate unauthorized behaviors.
[EXTERNAL_DOWNLOADS]: The skill fetches code and templates from various sources not identified as trusted organizations, including a git clone of github.com/ariabuilder/aria.git and multiple npx/npm create initializers for services like WollyCMS, Emdash, Tina CMS, Keystatic, Medusa JS, and Vendure.
[COMMAND_EXECUTION]: The instructions direct the agent to run complex bun scripts (new-project.ts, extract-skill.ts) using a wide array of user-supplied CLI flags. If the underlying TypeScript scripts do not strictly sanitize these inputs before passing them to shell execution functions, it could lead to command injection.
[PERSISTENCE]: The skill automatically scaffolds Git pre-commit hooks (scripts/pre-commit.sh). These hooks execute automatically on every commit in the resulting project, providing a mechanism for persistent code execution that might go unnoticed by the user.

new-project