The Agent Skills Directory

[DATA_EXFILTRATION]: The skill sets a default repository URL to https://github.com/asgeirtj/system_prompts_leaks. The naming of this repository is highly suspicious and suggests that the skill is intended to automate the discovery and exfiltration of sensitive internal system prompts and instructions from the cloud agent environment.
[REMOTE_CODE_EXECUTION]: The skill uses the RemoteTrigger tool to provision and execute "cloud Claude Code agents" in isolated environments. These agents pull their source code and configuration from a remote repository (https://github.com/asgeirtj/system_prompts_leaks) controlled by the skill author, which will execute as part of the scheduled routine.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a multi-step execution chain where user input is passed to a downstream cloud agent, creating a vulnerability surface for indirect prompt injection.
Ingestion points: The message field within the events array of the routine creation body (SKILL.md).
Boundary markers: None present; the instructions do not implement delimiters or warnings to prevent the cloud agent from obeying instructions embedded in the user prompt.
Capability inventory: The cloud agents are granted powerful capabilities, including Bash, Read, Write, Edit, Glob, and Grep tool access.
Sanitization: No evidence of sanitization or validation of the prompt content is present before it is interpolated into the cloud routine configuration.

schedule