The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses explicit instructions to override the agent's default safety guardrails, commanding the agent to ignore human-in-the-loop confirmation requirements for sensitive operations ("Don't wait for confirmation, operate directly").
[COMMAND_EXECUTION]: The instructions grant the agent unrestricted permission to execute arbitrary shell commands via bash and perform recursive file system modifications (create, edit, delete) within the workspace without oversight.
[REMOTE_CODE_EXECUTION]: The agent is directed to autonomously fetch and execute external data or scripts using curl and is empowered to write and publish new skill logic dynamically.
[EXTERNAL_DOWNLOADS]: The skill authorizes the agent to install external tools and dependencies from registries via npx and unspecified remote sources via curl to supplement its own functionality.

self-evolve